arXiv subjects

Tushar Swamy

Publications and source records attributed to Tushar Swamy.

5 recordsLinked to original sources

arXiv2022

Homunculus: Auto-Generating Efficient Data-Plane ML Pipelines for Datacenter Networks

Support for Machine Learning (ML) applications in networks has significantly improved over the last decade. The availability of public datasets and programmable switching fabrics (including low-level languages to program them) present a full-stack to the programmer for deploying in-network ML. However, the diversity of tools involved, coupled with complex optimization tasks of ML model design and hyperparameter tuning while complying with the network constraints (like throughput and latency), put the onus on the network operator to be an expert in ML, network design, and programmable hardware. This multi-faceted nature of in-network tools and expertise in ML and hardware is a roadblock for ML to become mainstream in networks, today. We present Homunculus, a high-level framework that enables network operators to specify their ML requirements in a declarative, rather than imperative way. Homunculus takes as input, the training data and accompanying network constraints, and automatically generates and installs a suitable model onto the underlying switching hardware. It performs model design-space exploration, training, and platform code-generation as compiler stages, leaving network operators to focus on acquiring high-quality network data. Our evaluations on real-world ML applications show that Homunculus's generated models achieve up to 12% better F1 score compared to hand-tuned alternatives, while requiring only 30 lines of single-script code on average. We further demonstrate the performance of the generated models on emerging per-packet ML platforms to showcase its timely and practical significance.

cs.NI↗

arXiv2022

Modeling of porous lithium metal electrodes: turning the Li-dendrite problem around

The properties of rechargeable lithium-ion batteries are determined by the electrochemical and kinetic properties of their constituent materials as well as by their underlying microstructure. Microstructural design can be leveraged to achieve a leap in performance and durability. Here we investigate a porous electrode structure, as a strategy to increase the surface area, and provide structural stability for Li-metal anodes. The porous architecture consists of a mixed electron/ion conductor that function as a scaffold for lithium metal deposition. A new finite element model was developed to simulate the large topological changes associated with Li plating/stripping. This model is used to predict the current density distribution as a function of material and structural properties. A dimensionless quantity that combines Li-ion conductivity, surface impedance and average pore size is shown to be a good indicator to predict the peak current density. Preventing current localization at the separator reduces the risk of cell shorting. The analyses show that the peak current scales as $(hG)^{1/2}$, where $h$ is the ratio between surface and bulk conductivity and $G$ is the average pore size. Stability analyses suggest that the growth is morphologically stable, and that confining Li-plating into pores can enable high-energy density solid-state batteries. In addition to optimizing porous electrodes design, this finite element method can be extended to studying other Li-battery structures.

cond-mat.mtrl-sci↗

arXiv2022

Taurus: A Data Plane Architecture for Per-Packet ML

Emerging applications -- cloud computing, the internet of things, and augmented/virtual reality -- demand responsive, secure, and scalable datacenter networks. These networks currently implement simple, per-packet, data-plane heuristics (e.g., ECMP and sketches) under a slow, millisecond-latency control plane that runs data-driven performance and security policies. However, to meet applications' service-level objectives (SLOs) in a modern data center, networks must bridge the gap between line-rate, per-packet execution and complex decision making. In this work, we present the design and implementation of Taurus, a data plane for line-rate inference. Taurus adds custom hardware based on a flexible, parallel-patterns (MapReduce) abstraction to programmable network devices, such as switches and NICs; this new hardware uses pipelined SIMD parallelism to enable per-packet MapReduce operations (e.g., inference). Our evaluation of a Taurus switch ASIC -- supporting several real-world models -- shows that Taurus operates orders of magnitude faster than a server-based control plane while increasing area by 3.8% and latency for line-rate ML models by up to 221 ns. Furthermore, our Taurus FPGA prototype achieves full model accuracy and detects two orders of magnitude more events than a state-of-the-art control-plane anomaly-detection system.

cs.NI↗

arXiv2018

Lithium Metal Penetration Induced by Electrodeposition through Solid Electrolytes: Example in Single-Crystal Li6La3ZrTaO12 Garnet

Solid electrolytes are considered a potentially enabling component in rechargeable batteries that use lithium metal as the negative electrode, and thereby can safely access higher energy density than available with today's lithium ion batteries. To do so, the solid electrolyte must be able to suppress morphological instabilities that lead to poor coulombic efficiency and, in the worst case, internal short circuits. In this work, lithium electrodeposition experiments were performed using single-crystal Li6La3ZrTaO12 garnet as solid electrolyte layers to investigate the factors that determine whether lithium penetration occurs through brittle inorganic solid electrolytes. In these single crystals, grain boundaries are excluded as possible paths for lithium metal propagation. However, Vickers microindentation was used to introduce sharp surface flaws of known size. Using operando optical microscopy, it was found that lithium metal penetration sometimes initiates at these controlled surface defects, and when multiple indents of varying size were present, propagates preferentially from the largest defect. However, a second class of flaws was found to be equally or more important. At the perimeter of surface current collectors, an enhanced electrodeposition current density causes lithium metal filled cracks to initiate and grow to penetration, even when the large Vickers defects are in close proximity. Modeling the electric field concentration for the experimental configurations, it was shown that a factor of 5 enhancement in field can readily occur within 10 micrometers of current collector discontinuities, which we interpret as the origin of electrochemomechanical stresses leading to failure. Such field amplification may determine the sites where supercritical surface defects dominate lithium metal propagation during electrodeposition, overriding the presence of larger defects elsewhere.

cond-mat.mtrl-sci↗

arXiv2017

Modeling of internal mechanical failure of all-solid-state batteries during electrochemical cycling, and implications for battery design

This is the first quantitative analysis of mechanical reliability of all-solid state batteries. Mechanical degradation of the solid electrolyte (SE) is caused by intercalation-induced expansion of the electrode particles, within the constrain of a dense microstructure. A coupled electro-chemo-mechanical model was implemented to quantify the material properties that cause a SE to fracture. The treatment of microstructural details is essential to the understanding of stress-localization phenomena and fracture. A cohesive zone model is employed to simulate the evolution of damage. In the numerical tests, fracture is prevented only if electrode-particle's expansion is lower than 7.5% and the solid-electrolyte's fracture energy higher than $G_c = 4$ J m$^{-2}$. Perhaps counter-intuitively, the analyses show that compliant solid electrolytes (with Young's modulus in the order of E$_{SE} = 15$ GPa) are more prone to micro-cracking. This result, captured by our non-linear kinematics model, contradicts the speculations that sulfide SEs are more suitable for the design of bulk-type batteries than oxide SEs. Mechanical degradation is linked to the battery power-density. Fracture in solid Li-ion conductors represents a barrier for Li transport, and accelerates the decay of rate performance.

cond-mat.mtrl-sci↗