Searcharxiv⌕ Search

arXiv subjects

Yuan Yu

Publications and source records attributed to Yuan Yu.

28 records · Page 2Linked to original sources
arXiv2022

Dynamic doping and Cottrell atmosphere optimize the thermoelectric performance of n-type PbTe

High thermoelectric energy conversion efficiency requires a large figure-of-merit, zT, over a broad temperature range. To achieve this, we optimize the carrier concentrations of n-type PbTe from room up to hot-end temperatures by co-doping Bi and Ag. Bi is an efficient n-type dopant in PbTe, often leading to excessive carrier concentration at room temperature. As revealed by density functional theory calculations, the formation of Bi and Ag defect complexes is exploited to optimize the room temperature carrier concentration. At elevated temperatures, we demonstrate the dynamic dissolution of Ag2Te precipitates in PbTe in situ by heating in a scanning transmission electron microscope. The release of n-type Ag interstitials with increasing temperature fulfills the requirement of higher carrier concentrations at the hot end. Moreover, as characterized by atom probe tomography, Ag atoms aggregate along parallel dislocation arrays to form Cottrell atmospheres. This results in enhanced phonon scattering and leads to a low lattice thermal conductivity. As a result of the synergy of dynamic doping and phonon scattering at decorated dislocations, an average zT of 1.0 is achieved in n-type Bi/Ag-codoped PbTe between 400 and 825 K. Introducing dopants with temperature-dependent solubility and strong interaction with dislocation cores enables simultaneous optimization of the average power factor and thermal conductivity, providing a new concept to exploit in the field of thermoelectrics.

cond-mat.mtrl-sci↗
arXiv2021

Dopant-segregation to grain boundaries controls electrical conductivity of n-type NbCo(Pt)Sn half-Heusler alloy mediating thermoelectric performance

Science-driven design of future thermoelectric materials requires a deep understanding of the fundamental relationships between microstructure and transport properties. Grain boundaries in polycrystalline materials influence the thermoelectric performance through the scattering of phonons or the trapping of electrons due to space-charge effects. Yet, the current lack of careful investigations on grain boundary-associated features hinders further optimization of properties. Here, we study n-type NbCo1-xPtxSn half-Heusler alloys, which were synthesized by ball milling and spark plasma sintering (SPS). Post-SPS annealing was performed on one sample, leading to improved low-temperature electrical conductivity. The microstructure of both samples was examined by electron microscopy and atom probe tomography. The grain size increases from ~230 nm to ~2.38 μm upon annealing. Pt is found within grains and at grain boundaries, where it locally reduces the resistivity, as assessed by in situ four-point-probe electrical conductivity measurement. Our work showcases the correlation between microstructure and electrical conductivity, providing opportunities for future microstructural optimization by tuning the chemical composition at grain boundaries.

cond-mat.mtrl-sci↗
arXiv2019

Extending Relational Query Processing with ML Inference

The broadening adoption of machine learning in the enterprise is increasing the pressure for strict governance and cost-effective performance, in particular for the common and consequential steps of model storage and inference. The RDBMS provides a natural starting point, given its mature infrastructure for fast data access and processing, along with support for enterprise features (e.g., encryption, auditing, high-availability). To take advantage of all of the above, we need to address a key concern: Can in-RDBMS scoring of ML models match (outperform?) the performance of dedicated frameworks? We answer the above positively by building Raven, a system that leverages native integration of ML runtimes (i.e., ONNX Runtime) deep within SQL Server, and a unified intermediate representation (IR) to enable advanced cross-optimizations between ML and DB operators. In this optimization space, we discover the most exciting research opportunities that combine DB/Compiler/ML thinking. Our initial evaluation on real data demonstrates performance gains of up to 5.5x from the native integration of ML in SQL Server, and up to 24x from cross-optimizations--we will demonstrate Raven live during the conference talk.

cs.DB↗
arXiv2019

An optical all-pass filter realized by using self-compensation of loss

Filters are key devices in optical systems and are usually applied to signal suppression or selection based on their amplitude frequency responses. Different from amplitude filtering, an all-pass filter (APF) is a unique type of filter that exhibits a uniform amplitude and variable phase in the frequency response. Traditionally, an APF is designed to be ideal and lossless, which is impossible in practice. Here we demonstrate a method of realizing an APF even when transmission loss exists, and a possible structure for realizing the lossy APF is proposed. The method indicates that a lossy APF can be realized by introducing self-compensation of loss to the structure of an ideal single-stage APF, and an APF based on silicon-on-insulator (SOI) foundry is fabricated and demonstrated. The method of self-compensation of loss can also be used to realize APFs in the electrical domain. We anticipate that our work will inspire realizations of more APF-related devices that once seemed impossible in practice.

physics.optics↗
arXiv2018

A versatile lattice Boltzmann model for immiscible ternary fluid flows

We propose a lattice Boltzmann color-gradient model for immiscible ternary fluid flows, which is applicable to the fluids with a full range of interfacial tensions, especially in near-critical and critical states. An interfacial force for N-phase systems is derived based on the previously developed perturbation operator and is then introduced into the model using a body force scheme, which helps reduce spurious velocities. A generalized recoloring algorithm is applied to produce phase segregation and ensure immiscibility of three different fluids, where a novel form of segregation parameters is proposed by considering the existence of Neumann's triangle and the effect of equilibrium contact angle in three-phase junction. The proposed model is first validated with three typical examples, namely the interface capturing for two separate static droplets, the Young-Laplace test for a compound droplet, and the spreading of a droplet between two stratified fluids. This model is then used to study the structure and stability of double droplets in a static matrix. Consistent with the theoretical stability diagram, seven possible equilibrium morphologies are successfully reproduced by adjusting two ratios of the interfacial tensions. By simulating Janus droplets in various geometric configurations, the model is shown to be accurate when three interfacial tensions satisfy a Neumann's triangle. In addition, we also simulate the near-critical and critical states of double droplets where the outcomes are very sensitive to the model accuracy. Our results show that the present model is advantageous to three-phase flow simulations, and allows for accurate simulation of near-critical and critical states.

physics.comp-ph↗
arXiv2018

Dynamic Control Flow in Large-Scale Machine Learning

Many recent machine learning models rely on fine-grained dynamic control flow for training and inference. In particular, models based on recurrent neural networks and on reinforcement learning depend on recurrence relations, data-dependent conditional execution, and other features that call for dynamic control flow. These applications benefit from the ability to make rapid control-flow decisions across a set of computing devices in a distributed system. For performance, scalability, and expressiveness, a machine learning system must support dynamic control flow in distributed and heterogeneous environments. This paper presents a programming model for distributed machine learning that supports dynamic control flow. We describe the design of the programming model, and its implementation in TensorFlow, a distributed machine learning system. Our approach extends the use of dataflow graphs to represent machine learning models, offering several distinctive features. First, the branches of conditionals and bodies of loops can be partitioned across many machines to run on a set of heterogeneous devices, including CPUs, GPUs, and custom ASICs. Second, programs written in our model support automatic differentiation and distributed gradient computations, which are necessary for training machine learning models that use control flow. Third, our choice of non-strict semantics enables multiple loop iterations to execute in parallel across machines, and to overlap compute and I/O operations. We have done our work in the context of TensorFlow, and it has been used extensively in research and production. We evaluate it using several real-world applications, and demonstrate its performance and scalability.

cs.DC↗
arXiv2017

Lattice Boltzmann simulation of viscous fingering of immiscible displacement in a channel using an improved wetting scheme

An improved wetting boundary implementation strategy is proposed based on lattice Boltzmann color-gradient model in this paper. In this strategy, an extra interface force condition is demonstrated based on the diffuse interface assumption and is employed in contact line region. It has been validated by three benchmark problems: static droplet wetting on a flat surface and a curved surface, and dynamic capillary filling. Good performances are shown in all three cases. Relied on the strict validation to our scheme, the viscous fingering phenomenon of immiscible fluids displacement in a two-dimensional channel has been restudied in this paper. High viscosity ratio, wide range contact angle, accurate moving contact line and mutual independence between surface tension and viscosity are the obvious advantages of our model. We find the linear relationship between the contact angle and displacement velocity or variation of finger length. When the viscosity ratio is smaller than 20, the displacement velocity is increasing with increasing viscosity ratio and reducing capillary number, and when the viscosity ratio is larger than 20, the displacement velocity tends to a specific constant. A similar conclusion is obtained on the variation of finger length.

physics.flu-dyn↗
arXiv2016

TensorFlow: A system for large-scale machine learning

TensorFlow is a machine learning system that operates at large scale and in heterogeneous environments. TensorFlow uses dataflow graphs to represent computation, shared state, and the operations that mutate that state. It maps the nodes of a dataflow graph across many machines in a cluster, and within a machine across multiple computational devices, including multicore CPUs, general-purpose GPUs, and custom designed ASICs known as Tensor Processing Units (TPUs). This architecture gives flexibility to the application developer: whereas in previous "parameter server" designs the management of shared state is built into the system, TensorFlow enables developers to experiment with novel optimizations and training algorithms. TensorFlow supports a variety of applications, with particularly strong support for training and inference on deep neural networks. Several Google services use TensorFlow in production, we have released it as an open-source project, and it has become widely used for machine learning research. In this paper, we describe the TensorFlow dataflow model in contrast to existing systems, and demonstrate the compelling performance that TensorFlow achieves for several real-world applications.

cs.DC↗
arXiv2016

TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems

TensorFlow is an interface for expressing machine learning algorithms, and an implementation for executing such algorithms. A computation expressed using TensorFlow can be executed with little or no change on a wide variety of heterogeneous systems, ranging from mobile devices such as phones and tablets up to large-scale distributed systems of hundreds of machines and thousands of computational devices such as GPU cards. The system is flexible and can be used to express a wide variety of algorithms, including training and inference algorithms for deep neural network models, and it has been used for conducting research and for deploying machine learning systems into production across more than a dozen areas of computer science and other fields, including speech recognition, computer vision, robotics, information retrieval, natural language processing, geographic information extraction, and computational drug discovery. This paper describes the TensorFlow interface and an implementation of that interface that we have built at Google. The TensorFlow API and a reference implementation were released as an open-source package under the Apache 2.0 license in November, 2015 and are available at www.tensorflow.org.

cs.DC↗