arXiv subjects

David Jiménez

Publications and source records attributed to David Jiménez.

At least 19 recordsLinked to original sources

arXiv2025

Large Language Models Meet Extreme Multi-label Classification: Scaling and Multi-modal Framework

Foundation models have revolutionized artificial intelligence across numerous domains, yet their transformative potential remains largely untapped in Extreme Multi-label Classification (XMC). Queries in XMC are associated with relevant labels from extremely large label spaces, where it is critical to strike a balance between efficiency and performance. Therefore, many recent approaches efficiently pose XMC as a maximum inner product search between embeddings learned from small encoder-only transformer architectures. In this paper, we address two important aspects in XMC: how to effectively harness larger decoder-only models, and how to exploit visual information while maintaining computational efficiency. We demonstrate that both play a critical role in XMC separately and can be combined for improved performance. We show that a few billion-size decoder can deliver substantial improvements while keeping computational overhead manageable. Furthermore, our Vision-enhanced eXtreme Multi-label Learning framework (ViXML) efficiently integrates foundation vision models by pooling a single embedding per image. This limits computational growth while unlocking multi-modal capabilities. Remarkably, ViXML with small encoders outperforms text-only decoder in most cases, showing that an image is worth billions of parameters. Finally, we present an extension of existing text-only datasets to exploit visual metadata and make them available for future benchmarking. Comprehensive experiments across four public text-only datasets and their corresponding image enhanced versions validate our proposals' effectiveness, surpassing previous state-of-the-art by up to +8.21\% in P@1 on the largest dataset. ViXML's code is available at https://github.com/DiegoOrtego/vixml.

cs.CV↗

arXiv2025

A Taylor Series Approximation Model for Characterizing the Output Resistance of a GFET

The mobility-degradation-based model for the drain-to-source or output resistance of a graphene field-effect-transistor is linearized here using a Taylor series approximation. This simplification is shown to be valid from magnitudes of the gate voltage not significantly higher than the Dirac voltage, and it enables the analytical determination of the transconductance parameter, the voltage related to residual charges, and a bias-independent series resistance of the GFET. Furthermore, a continuous representation of the device's static response is achieved when substituting the extracted parameters into the model, regardless the transfer characteristic symmetry with respect to the Dirac voltage.

cond-mat.mes-hall↗

arXiv2024

Prototypical Extreme Multi-label Classification with a Dynamic Margin Loss

Extreme Multi-label Classification (XMC) methods predict relevant labels for a given query in an extremely large label space. Recent works in XMC address this problem using deep encoders that project text descriptions to an embedding space suitable for recovering the closest labels. However, learning deep models can be computationally expensive in large output spaces, resulting in a trade-off between high performing brute-force approaches and efficient solutions. In this paper, we propose PRIME, a XMC method that employs a novel prototypical contrastive learning technique to reconcile efficiency and performance surpassing brute-force approaches. We frame XMC as a data-to-prototype prediction task where label prototypes aggregate information from related queries. More precisely, we use a shallow transformer encoder that we coin as Label Prototype Network, which enriches label representations by aggregating text-based embeddings, label centroids and learnable free vectors. We jointly train a deep encoder and the Label Prototype Network using an adaptive triplet loss objective that better adapts to the high granularity and ambiguity of extreme label spaces. PRIME achieves state-of-the-art results in several public benchmarks of different sizes and domains, while keeping the model efficient.

cs.LG↗

arXiv2024

A Scalable Compact Model for the Static Drain Current of Graphene FETs

The main target of this article is to propose for the first time a physics-based continuous and symmetric compact model that accurately captures IV experimental dependencies induced by geometrical scaling effects for graphene transistor (GFET) technologies. Such a scalable model is an indispensable ingredient for the boost of large-scale GFET applications, as it has been already proved in solid industry-based CMOS technologies. Dependencies of the physical model parameters on channel dimensions, are thoroughly investigated, and semi?empirical expressions are derived, which precisely characterize such behaviors for an industry-based GFET technology, as well as for others developed in research labs. This work aims at the establishment of the first industry standard GFET compact model that can be integrated in circuit simulation tools and hence, can contribute to the update of GFET technology from the research level to massive industry production.

physics.app-ph↗

arXiv2023

Exploiting ambipolarity in graphene field-effect transistors for novel designs on high-frequency analog electronics

Exploiting ambipolar electrical conductivity based on graphene field-effect transistors has raised enormous interest for high-frequency (HF) analog electronics. Controlling the device polarity, by biasing the graphene transistor around the vertex of the V-shaped transfer curve, enables to redesign and highly simplify conventional analog circuits, and simultaneously to seek for multifunctionalities specially in the HF domain. We present, here, new insights for the design of different HF applications such as power amplifiers, mixers, frequency multipliers, phase shifters, and modulators that specifically leverage the inherent ambipolarity of graphene-based transistors.

cond-mat.mes-hall↗

arXiv2023

Characterization of the Intrinsic and Extrinsic Resistances of a Microwave Graphene FET Under Zero Transconductance Conditions

Graphene field-effect transistors exhibit negligible transconductance under two scenarios: for any gate-to-source voltage when the drain-to-source voltage is set to zero, and for an arbitrary drain-to-source voltage provided that the gate-to-source voltage equals the Dirac voltage. Hence, extracting the channel and the parasitic series resistances from S-parameters under these conditions enables analyzing their dependence on the gate and drain biases. This is fundamental to assess the portion of the output resistance that is controlled by the gate. Besides, the drain bias dependence of the drain and source resistances is also evidenced. Within the proposal, resistive components accounting for the lossy nature of the gate capacitance are incorporated into the model, which exhibits a broadband correlation with experimental data. This avoids the series resistances to be considered as frequency dependent in the model.

cond-mat.mes-hall↗

arXiv2023

LEA: Improving Sentence Similarity Robustness to Typos Using Lexical Attention Bias

Textual noise, such as typos or abbreviations, is a well-known issue that penalizes vanilla Transformers for most downstream tasks. We show that this is also the case for sentence similarity, a fundamental task in multiple domains, e.g. matching, retrieval or paraphrasing. Sentence similarity can be approached using cross-encoders, where the two sentences are concatenated in the input allowing the model to exploit the inter-relations between them. Previous works addressing the noise issue mainly rely on data augmentation strategies, showing improved robustness when dealing with corrupted samples that are similar to the ones used for training. However, all these methods still suffer from the token distribution shift induced by typos. In this work, we propose to tackle textual noise by equipping cross-encoders with a novel LExical-aware Attention module (LEA) that incorporates lexical similarities between words in both sentences. By using raw text similarities, our approach avoids the tokenization shift problem obtaining improved robustness. We demonstrate that the attention bias introduced by LEA helps cross-encoders to tackle complex scenarios with textual noise, specially in domains with short-text descriptions and limited context. Experiments using three popular Transformer encoders in five e-commerce datasets for product matching show that LEA consistently boosts performance under the presence of noise, while remaining competitive on the original (clean) splits. We also evaluate our approach in two datasets for textual entailment and paraphrasing showing that LEA is robust to typos in domains with longer sentences and more natural context. Additionally, we thoroughly analyze several design choices in our approach, providing insights about the impact of the decisions made and fostering future research in cross-encoders dealing with typos.

cs.CL↗

arXiv2023

Static and small-signal modeling of radiofrequency hexagonal boron nitride switches

A first modeling approximation of the general performance of radiofrequency (RF) switches based on hexagonal boron nitride (hBN), a two-dimensional (2D) dielectric material is presented. The I-V characteristics intrinsic and extrinsic impedance parameters, the return loss, insertion loss and isolation of RF 2D switches fabricated with hBN are described here by a equivalent circuit models. Straightforward analytical expressions are obtained. In contrast to conventional switches, the unique RF performance of the hBN switch, at ON-state, i.e., a direct improvement with frequency of the insertion loss, is accurately described by considering a capacitor in the intrinsic part of the model. The latter is suggested to be related to storaged charge during the resistive switching mechanism. The highest mean relative error obtained between modeling and measurements of the return loss is of 7.6% with the approach presented here which overcomes the 42.5% of difference obtained with a previous model with an incomplete intrinsic device description.

physics.app-ph↗

arXiv2023

Physics-based bias-dependent compact modeling of 1/f noise in single- to few- layer 2D-FETs

1/f noise is a critical figure of merit for the performance of transistors and circuits. For two-dimensional devices (2D-FETs), and especially for applications in the GHz range where short-channel FETs are required, velocity saturation (VS) effect can result in the reduction of 1/f noise at high longitudinal electric fields. A new physics-based compact model is for the first time introduced for single- to few- layer 2D-FETs in this study, precisely validating 1/f noise experiments for various types of devices. The proposed model mainly accounts for the measured 1/f noise bias dependence as the latter is defined by different physical mechanisms. Thus, analytical expressions are derived, valid in all regions of operation in contrast to conventional approaches available in literature so far, accounting for carrier number fluctuation (DN), mobility fluctuation (Dmu}) and contact resistance (DR) effects based on the underlying physics that rules these devices. DN mechanism due to trapping/detrapping together with an intense Coulomb scattering effect, dominates 1/f noise from medium to strong accumulation region while Dmu, is also demonstrated to modestly contribute in subthreshold region. DR can also be significant in very high carrier density. The VS induced reduction of 1/f noise measurements at high electric fields, is also remarkably captured by the model. The physical validity of the model can also assist in extracting credible conclusions when conducting comparisons between experimental data from devices with different materials or dielectrics.

cond-mat.mes-hall↗

arXiv2023

Straightforward Bias and Frequency Dependent Small-Signal Model Extraction for Single-Layer Graphene FETs

We propose an explicit small-signal graphene field-effect transistor (GFET) parameter extraction procedure based on a charge-based quasi-static model. The dependence of the small-signal parameters on both gate voltage and frequency is precisely validated by high-frequency (up to 18 GHz) on-wafer measurements from a 300 nm device. These parameters are studied simultaneously, in contrast to other works which focus exclusively on few. Efficient procedures have been applied to GFETs for the first time to remove contact and gate resistances from the Y-parameters. The use of these methods yields straightforward equations for extracting the small-signal model parameters, which is extremely useful for radio-frequency circuit design. Furthermore, we show for the first time experimental validation vs. both gate voltage and frequency of the intrinsic GFET non-reciprocal capacitance model. Accurate models are also presented for the gate voltage-dependence of the measured unity-gain and maximum oscillation frequencies as well as of the current and power gains.

cond-mat.mes-hall↗

arXiv2022

Compact modeling technology for the simulation of integrated circuits based on graphene field-effect transistors

In this study, we report the progress made towards the definition of a modular compact modeling technology for graphene field-effect transistors (GFET) that enables the electrical analysis of arbitrary GFET-based integrated circuits. A set of primary models embracing the main physical principles defines the ideal GFET response under DC, transient (time domain), AC (frequency domain), and noise (frequency domain) analysis. Other set of secondary models accounts for the GFET non-idealities, such as extrinsic-, short-channel-, trapping/detrapping-, self-heating-, and non-quasi static-effects, which could have a significant impact under static and/or dynamic operation. At both device and circuit levels, significant consistency is demonstrated between the simulation output and experimental data for relevant operating conditions. Additionally, we provide a perspective of the challenges during the scale up of the GFET modeling technology towards higher technology readiness levels while drawing a collaborative scenario among fabrication technology groups, modeling groups, and circuit designers.

physics.app-ph↗

arXiv2022

Schottky-like barrier characterization of field-effect transistors with multiple quasi-ballistic channels

The potential barrier height at the interface formed by a metal contact and multiple one-dimensional (1D) quasi-ballistic channels in field-effect transistors (FETs) is evaluated across different carbon nanotube and nanowire device technologies by means of a Landauer-Büttiker-based extraction methodology (LBM) adapted for multiple 1D-channels. The extraction methodology yields values for an effective Schottky barrier height and a gate coupling coefficient, an indicator of the device working at the quantum capacitance limit. The novel LBM-based approach embracing the mechanisms in 1D electronics is compared to the conventional activation energy method not considering such effects. The latter approach underestimates the potential barrier height at metal-channel interfaces in comparison to the novel methodology. A test structure based on a displaced gate device is proposed based on numerical device simulation results towards an improved accuracy of the method.

cond-mat.mes-hall↗

arXiv2022

Block-SCL: Blocking Matters for Supervised Contrastive Learning in Product Matching

Product matching is a fundamental step for the global understanding of consumer behavior in e-commerce. In practice, product matching refers to the task of deciding if two product offers from different data sources (e.g. retailers) represent the same product. Standard pipelines use a previous stage called blocking, where for a given product offer a set of potential matching candidates are retrieved based on similar characteristics (e.g. same brand, category, flavor, etc.). From these similar product candidates, those that are not a match can be considered hard negatives. We present Block-SCL, a strategy that uses the blocking output to make the most of Supervised Contrastive Learning (SCL). Concretely, Block-SCL builds enriched batches using the hard-negatives samples obtained in the blocking stage. These batches provide a strong training signal leading the model to learn more meaningful sentence embeddings for product matching. Experimental results in several public datasets demonstrate that Block-SCL achieves state-of-the-art results despite only using short product titles as input, no data augmentation, and a lighter transformer backbone than competing methods.

cs.CL↗

arXiv2022

An extraction method for mobility degradation and contact resistance of graphene transistors

The intrinsic mobility degradation coefficient, contact resistance and the transconductance parameter of graphene field-effect transistors (GFETs) are extracted for different technologies by considering a novel transport model embracing mobility degradation effects within the charge channel control description. By considering the mobility degradation-based model, a straightforward extraction methodology, not provided before, is enabled by applying the concept of the well-known Y-function to the \textit{I-V} device characteristics. The method works regardless the gate device architecture. An accurate description of experimental data of fabricated devices is achieved with the underlying transport equation by using the extracted parameters. An evaluation of the channel resistance, enabled by the extracted parameters here, has been also provided.

cond-mat.mes-hall↗

arXiv2021

The gate tunable 2D pn junction driven out-of-equilibrium

We have investigated the electrostatics and electronic transport of the gate tunable 2D pn junction by implementing a comprehensive physics-based simulator that self-consistently solves the 2D Poisson's equation coupled to the drift-diffusion current and continuity equations. The simulator considers the strong influence of the out-of-plane electric field through the gate dielectric and the presence of interface states. The impact of parameters such as gate capacitance, energy gap and interface trap states density have been considered to model properties such as the depletion width, rectification factor and depletion and diffusion capacitances. The present work opens the door to a wider exploration of potential advantages that gate tunable 2D pn junctions could bring in terms of figures of merit.

cond-mat.mes-hall↗

arXiv2021

Compact Modeling of pH-Sensitive FETs Based on Two-Dimensional Semiconductors

We present a physics-based circuit-compatible model for pH-sensitive field-effect transistors based on two-dimensional (2D) materials. The electrostatics along the electrolyte-gated 2D-semiconductor stack is treated by solving the Poisson equation including the Site-Binding model and the Gouy-Chapman-Stern approach, while the carrier transport is described by the drift-diffusion theory. The proposed model is provided in an analytical form and then implemented in Verilog-A, making it compatible with standard technology computer-aided design tools employed for circuit simulation. The model is benchmarked against two experimental transition-metal-dichalcogenide (MoS2 and ReS2) based ion sensors, showing excellent agreement when predicting the drain current, threshold voltage shift, and current/voltage sensitivity measurements for different pH concentrations.

physics.app-ph↗

arXiv2021

Unveiling the impact of the bias-dependent charge neutrality point on graphene-based multi-transistor applications

The Dirac voltage of a graphene field-effect transistor (GFET) stands for the gate bias that sets the charge neutrality condition in the channel, thus resulting in a minimum conductivity. Controlling its dependence on the terminal biases is crucial for the design and optimization of radio-frequency applications based on multiple GFETs. However, the previous analysis of such dependence carried out for a single device can lead to confusion and if not properly understood could result in circuit designs with poor performance. The control of the Dirac point shift (DPS) is particularly important for the deployment of graphene-based differential circuit topologies where keeping a strict symmetry between the electrical balanced branches is crucial for exploiting the advantages of such topologies. This note sheds light on the impact of terminal biases on the DPS in a real device and sets a rigorous methodology to control it so to eventually optimize and exploit the performance of radio-frequency applications based on GFETs.

physics.app-ph↗

arXiv2020

A small-signal GFET equivalent circuit considering an explicit contribution of contact resistances

A small-signal equivalent circuit for graphene field-effect transistors is proposed considering the explicit contribution of effects at the metal-graphene interfaces by means of contact resistances. A methodology to separate the contact resistances from intrinsic parameters, obtained by a de-embedding process, and extrinsic parameters of the circuit is considered. The experimental high-frequency performance of three devices from two different GFET technologies is properly described by the proposed small-signal circuit. Some model parameters scale with the device footprint. The correct detachment of contact resistances from the internal transistor enables to assess their impact on the intrinsic cutoff frequency of the studied devices.

physics.app-ph↗