arXiv subjects

Piyush Kumar

Publications and source records attributed to Piyush Kumar.

At least 37 records · Page 2Linked to original sources

arXiv2024

Surface Structuring of Patterned 4H-SiC Surfaces Using a SiC/Si/SiC Sandwich Approach

Mesa- and trench-patterned surfaces of 4H-SiC(0001) 4{\textdegree}off wafers were structured in macrosteps using Si melting in a SiC-Si-SiC sandwich configuration. Si spreading difficulties were observed in the case of trench-patterned samples while the attempts on mesa-patterned ones were more successful. In the latter case, parallel macrosteps were formed on both the dry-etched and unetched areas though these macrosteps rarely cross the patterns edges. The proposed mechanism involved preferential etching at Si-C bilayer step edges and fast lateral propagation along the [1120] direction.

cond-mat.mtrl-sci↗

arXiv2024

A 5T-2MTJ STT-assisted Spin Orbit Torque based Ternary Content Addressable Memory for Hardware Accelerators

In this work, we present a novel non-volatile spin transfer torque (STT) assisted spin-orbit torque (SOT) based ternary content addressable memory (TCAM) with 5 transistors and 2 magnetic tunnel junctions (MTJs). We perform a comprehensive study of the proposed design from the device-level to application-level. At the device-level, various write characteristics such as write error rate, time, and current have been obtained using micromagnetic simulations. The array-level search and write performance have been evaluated based on SPICE circuit simulations with layout extracted parasitics for bitcells while also accounting for the impact of interconnect parasitics at the 7nm technology node. A search error rate of 3.9x10^-11 is projected for exact search while accounting for various sources of variation in the design. In addition, the resolution of the search operation is quantified under various scenarios to understand the achievable quality of the approximate search operations. Application-level performance and accuracy of the proposed design have been evaluated and benchmarked against other state-of-the-art CAM designs in the context of a CAM-based recommendation system.

cs.ET↗

arXiv2024

Empowering Abilities: Increasing Representation of Students with Disabilities in the STEM Field

The ExploreSTEM Summer Camps 2023 were designed to deliver inclusive STEM education to students aged 14 to 22 years with disabilities. This paper presents a thorough examination of the 2023 camp program, emphasizing the pivotal role of inclusive STEM education in potentially shaping students' personal and academic trajectories. The curriculum encompassed four weeklong fundamental STEM domains: Internet of Things (IoT), Computational Engineering, Artificial Intelligence (AI), and Augmented and Virtual Reality (AR/VR). Within Camp 1, students actively engaged with Dash robots, employing dedicated programming environments to command actions and gather sensor data, fostering interactions with the IoT platform and facilitating seamless data transmission. Camp 2 was dedicated to acquainting students with foundational computational engineering principles, establishing a robust framework for comprehending intricate engineering concepts. Camp 3 commenced with insightful presentations elucidating AI applications across multifaceted industries, including engineering, healthcare, and education, illuminating AI's pervasive influence on contemporary society. The primary aim of Camp 4 was to introduce students to the immersive domains of AR and VR, showcasing their applications beyond conventional STEM disciplines into everyday life experiences. The amalgamation of informative presentations, interactive activities, and a nurturing learning environment cultivated an engaging and enriching experience for all participants. By embracing inclusivity and harnessing innovative pedagogical approaches, the ExploreSTEM Summer Camps empowered students to explore, innovate, and excel within the dynamic realm of STEM education.

econ.GN↗

arXiv2024

ShieldGemma: Generative AI Content Moderation Based on Gemma

We present ShieldGemma, a comprehensive suite of LLM-based safety content moderation models built upon Gemma2. These models provide robust, state-of-the-art predictions of safety risks across key harm types (sexually explicit, dangerous content, harassment, hate speech) in both user input and LLM-generated output. By evaluating on both public and internal benchmarks, we demonstrate superior performance compared to existing models, such as Llama Guard (+10.8\% AU-PRC on public benchmarks) and WildCard (+4.3\%). Additionally, we present a novel LLM-based data curation pipeline, adaptable to a variety of safety-related tasks and beyond. We have shown strong generalization performance for model trained mainly on synthetic data. By releasing ShieldGemma, we provide a valuable resource to the research community, advancing LLM safety and enabling the creation of more effective content moderation solutions for developers.

cs.CL↗

arXiv2024

Malicious Path Manipulations via Exploitation of Representation Vulnerabilities of Vision-Language Navigation Systems

Building on the unprecedented capabilities of large language models for command understanding and zero-shot recognition of multi-modal vision-language transformers, visual language navigation (VLN) has emerged as an effective way to address multiple fundamental challenges toward a natural language interface to robot navigation. However, such vision-language models are inherently vulnerable due to the lack of semantic meaning of the underlying embedding space. Using a recently developed gradient based optimization procedure, we demonstrate that images can be modified imperceptibly to match the representation of totally different images and unrelated texts for a vision-language model. Building on this, we develop algorithms that can adversarially modify a minimal number of images so that the robot will follow a route of choice for commands that require a number of landmarks. We demonstrate that experimentally using a recently proposed VLN system; for a given navigation command, a robot can be made to follow drastically different routes. We also develop an efficient algorithm to detect such malicious modifications reliably based on the fact that the adversarially modified images have much higher sensitivity to added Gaussian noise than the original images.

cs.RO↗

arXiv2024

Machine Learning Models for Improved Tracking from Range-Doppler Map Images

Statistical tracking filters depend on accurate target measurements and uncertainty estimates for good tracking performance. In this work, we propose novel machine learning models for target detection and uncertainty estimation in range-Doppler map (RDM) images for Ground Moving Target Indicator (GMTI) radars. We show that by using the outputs of these models, we can significantly improve the performance of a multiple hypothesis tracker for complex multi-target air-to-ground tracking scenarios.

cs.CV↗

arXiv2024

Observation and manipulation of charge carrier distribution at the SiO$_2$/Si interface

Using low-energy muons, we map the charge carrier concentration as a function of depth and electric field across the \SiOSi interface up to a depth of \SI{100}{\nano\meter} in Si-based MOS capacitors. The results show that the formation of the anisotropic bond-centered muonium \MuBCz state in Si serves as a direct measure of the local changes in electronic structures. Different band-bending conditions could be distinguished, and the extension of the depletion width was directly extracted using the localized stopping and probing depth of the muons. Furthermore, electron build-up on the Si side of the \SiOO/Si interface, caused by the mirror charge induced by the fixed positive charge in the oxide and the image force effect, was observed. Our work represents a significant extension of the application of the muon spin rotation technique ($μ$SR) and lays the foundation for further research on direct observation of charge carrier density manipulation at technologically important semiconductor device interfaces.

cond-mat.other↗

arXiv2024

Cross-layer Modeling and Design of Content Addressable Memories in Advanced Technology Nodes for Similarity Search

In this paper we present a comprehensive design and benchmarking study of Content Addressable Memory (CAM) at the 7nm technology node in the context of similarity search applications. We design CAM cells based on SRAM, spin-orbit torque, and ferroelectric field effect transistor devices and from their layouts extract cell parasitics using state of the art EDA tools. These parasitics are used to develop SPICE netlists to model search operations. We use a CAM-based dataset search and a sequential recommendation system to highlight the application-level performance degradation due to interconnect parasitics. We propose and evaluate two solutions to mitigate interconnect effects.

cs.ET↗

arXiv2023

Relativistic dynamics of moving mirrors in CFT$_2$: quantum backreaction and black holes

There is a well-known correspondence between the physics of black hole evaporation and that of moving mirrors in QFT. However, most analyses in this subject rely on prescribed mirror trajectories. Here, we study the flat-space dynamics of $1+1$-dimensional Conformal Field Theories interacting with a relativistic boundary particle of mass $m$ acting as a perfect mirror. The trajectory of the latter is not fixed but follows its own relativistic equation of motion $F^μ=ma^μ$. For given initial conditions at past null infinity, we find the boundary particle's trajectory and the reflected energy-momentum of the quantum fields. For incoming vacuum states, the solution yields mirror orbits that correspond to extremal black holes. For the class of incoming states that produce orbits becoming null in finite proper time -- corresponding to the formation of a horizon -- at the classical level, the quantum backreaction avoids this endpoint rendering the mirror's velocity in lightcone coordinates finite. We investigate the behavior of the Averaged Null Energy Condition, which in this setup reduces to a boundary term.

hep-th↗

arXiv2023

Investigating Spontaneous SO(10) Symmetry Breaking in Type IIB Matrix Model

Non-perturbative formulations are essential to understand the dynamical compactification of extra dimensions in superstring theories. The type IIB (IKKT) matrix model in the large-$N$ limit is one such conjectured formulation for a ten-dimensional type IIB superstring. In this model, a smooth spacetime manifold is expected to emerge from the eigenvalues of the ten bosonic matrices. When this happens, the SO(10) symmetry in the Euclidean signature must be spontaneously broken. The Euclidean version has a severe sign problem since the Pfaffian obtained after integrating out the fermions is inherently complex. In recent years, the complex Langevin method (CLM) has successfully tackled the sign problem. We apply the CLM method to study the Euclidean version of the type IIB matrix model and investigate the possibility of spontaneous SO(10) symmetry breaking. In doing so, we encounter a singular-drift problem. To counter this, we introduce supersymmetry-preserving deformations with a Myers term. We study the spontaneous symmetry breaking in the original model at the vanishing deformation parameter limit. Our analysis indicates that the phase of the Pfaffian induces the spontaneous SO(10) symmetry breaking in the Euclidean type IIB model.

hep-lat↗

arXiv2023

On Regularization and Inference with Label Constraints

Prior knowledge and symbolic rules in machine learning are often expressed in the form of label constraints, especially in structured prediction problems. In this work, we compare two common strategies for encoding label constraints in a machine learning pipeline, regularization with constraints and constrained inference, by quantifying their impact on model performance. For regularization, we show that it narrows the generalization gap by precluding models that are inconsistent with the constraints. However, its preference for small violations introduces a bias toward a suboptimal model. For constrained inference, we show that it reduces the population risk by correcting a model's violation, and hence turns the violation into an advantage. Given these differences, we further explore the use of two approaches together and propose conditions for constrained inference to compensate for the bias introduced by regularization, aiming to improve both the model complexity and optimal risk.

cs.LG↗

arXiv2023

Safety and Fairness for Content Moderation in Generative Models

With significant advances in generative AI, new technologies are rapidly being deployed with generative components. Generative models are typically trained on large datasets, resulting in model behaviors that can mimic the worst of the content in the training data. Responsible deployment of generative technologies requires content moderation strategies, such as safety input and output filters. Here, we provide a theoretical framework for conceptualizing responsible content moderation of text-to-image generative technologies, including a demonstration of how to empirically measure the constructs we enumerate. We define and distinguish the concepts of safety, fairness, and metric equity, and enumerate example harms that can come in each domain. We then provide a demonstration of how the defined harms can be quantified. We conclude with a summary of how the style of harms quantification we demonstrate enables data-driven content moderation decisions.

cs.LG↗

arXiv2023

Salient Conditional Diffusion for Defending Against Backdoor Attacks

We propose a novel algorithm, Salient Conditional Diffusion (Sancdifi), a state-of-the-art defense against backdoor attacks. Sancdifi uses a denoising diffusion probabilistic model (DDPM) to degrade an image with noise and then recover said image using the learned reverse diffusion. Critically, we compute saliency map-based masks to condition our diffusion, allowing for stronger diffusion on the most salient pixels by the DDPM. As a result, Sancdifi is highly effective at diffusing out triggers in data poisoned by backdoor attacks. At the same time, it reliably recovers salient features when applied to clean data. This performance is achieved without requiring access to the model parameters of the Trojan network, meaning Sancdifi operates as a black-box defense.

cs.LG↗

arXiv2022

Investigation of the SiO2-SiC interface using low energy muon spin rotation spectroscopy

Using positive muons as local probes implanted at low energy enables gathering information about the material of interest with nanometer depth resolution (low energy muon spin rotation spectroscopy (LE-$μ$SR). In this work, we leverage the capabilities of LE-$μ$SR to perform an investigation of the SiO$_\text{2}$-SiC interface. Thermally oxidized samples are investigated before and after annealing in nitric oxide (NO) and argon (Ar) ambience. Thermal oxidation is found to result in structural changes both in the SiC crystal close to the interface and at the interface itself. Annealing in NO environment is known to passivate the defects leading to a reduction of the density of interface traps (D$_{it}$); LE-$μ$SR further reveals that the NO annealing results in a thin layer of high carrier concentration in SiC, extending to more than 50 nm depending on the annealing conditions. We also see indications of Si vacancy (V$_{Si}$) formation in SiC after thermal oxidation. Following NO annealing, nitrogen occupies the V$_{Si}$ sites, leading to the reduction in D$_{it}$ and at the same time, creating a charge-carrier-rich region near the interface. By comparing the LE-$μ$SR data from a sample with known doping density, we perform a high-resolution quantification of the free carrier concentration near the interface after NO annealing and discuss the origin of observed near-surface variations. Finally, the depletion of carriers in a MOS capacitor in the region below the interface is shown using LE-$μ$SR. The NO annealed sample shows the narrowest depletion region, likely due to the reduced D$_{it}$ and charge-carrier-rich region near the interface. Our findings demonstrate the many benefits of utilizing LE-$μ$SR to study critical regions of semiconductor devices that have been inaccessible with other techniques while retaining nanoscale depth resolution and a non-destructive approach.

cond-mat.mtrl-sci↗

arXiv2022

Complex Langevin study of spontaneous symmetry breaking in IKKT matrix model

The IKKT matrix model, in the large-$N$ limit, is conjectured to be a non-perturbative definition of the ten-dimensional type IIB superstring theory. In this work, we investigate the possibility of spontaneous breaking of the ten-dimensional rotational symmetry in the Euclidean IKKT model. Since the effective action, after integrating out the fermions, is inherently complex, we use the complex Langevin dynamics to study the model. In order to evade the singular-drift problem in the model, we add supersymmetry preserving deformations and then take the vanishing limit of the deformations. Our analysis suggests that the phase of the Pfaffian indeed induces the spontaneous SO(10) symmetry breaking in the Euclidean IKKT model.

hep-lat↗

arXiv2022

Defect Profiling of Oxide-Semiconductor Interfaces Using Low-Energy Muons

Muon spin rotation with low-energy muons (LEμSR) is a powerful nuclear method where electrical and magnetic properties of surface-near regions and thin films can be studied on a length scale of $\approx$\SI{200}{\nano\meter}. In this work, we show the potential of utilizing low-energy muons for a depth-resolved characterization of oxide-semiconductor interfaces, i.e. for silicon (Si) and silicon carbide (4H-SiC). Silicon dioxide (SiO$_2$) grown by plasma-enhanced chemical vapor deposition (PECVD) and by thermal oxidation of the SiO$_2$-semiconductor interface are compared with respect to interface and defect formation. The nanometer depth resolution of μallows for a clear distinction between the oxide and semiconductor layers, while also quantifying the extension of structural changes caused by the oxidation of both Si and SiC.

cond-mat.mtrl-sci↗

arXiv2021

Risk Bounds for Learning via Hilbert Coresets

We develop a formalism for constructing stochastic upper bounds on the expected full sample risk for supervised classification tasks via the Hilbert coresets approach within a transductive framework. We explicitly compute tight and meaningful bounds for complex datasets and complex hypothesis classes such as state-of-the-art deep neural network architectures. The bounds we develop exhibit nice properties: i) the bounds are non-uniform in the hypothesis space, ii) in many practical examples, the bounds become effectively deterministic by appropriate choice of prior and training data-dependent posterior distributions on the hypothesis space, and iii) the bounds become significantly better with increase in the size of the training set. We also lay out some ideas to explore for future research.

cs.LG↗

arXiv2020

Design, Analysis & Prototyping of a Semi-Automated Staircase-Climbing Rehabilitation Robot

In this paper, we describe the mechanical design, system overview, integration and control techniques associated with SKALA, a unique large-sized robot for carrying a person with physical disabilities, up and down staircases. As a regular wheelchair is unable to perform such a maneuver, the system functions as a non-conventional wheelchair with several intelligent features. We describe the unique mechanical design and the design choices associated with it. We showcase the embedded control architecture that allows for several different modes of teleoperation, all of which have been described in detail. We further investigate the architecture associated with the autonomous operation of the system.

cs.RO↗