<?xml version='1.0' encoding='UTF-8'?><?xml-stylesheet href='static/style.xsl' type='text/xsl'?><OAI-PMH xmlns="http://www.openarchives.org/OAI/2.0/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/ http://www.openarchives.org/OAI/2.0/OAI-PMH.xsd"><responseDate>2026-11-22T10:40:17Z</responseDate><request verb="GetRecord" identifier="oai:ecommons.cornell.edu:1813/117514" metadataPrefix="dim">https://ecommons.cornell.edu/server/oai/request</request><GetRecord><record><header><identifier>oai:ecommons.cornell.edu:1813/117514</identifier><datestamp>2026-05-15T17:54:24Z</datestamp><setSpec>com_1813_35</setSpec><setSpec>col_1813_47</setSpec></header><metadata><dim:dim xmlns:dim="http://www.dspace.org/xmlns/dspace/dim" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:doc="http://www.lyncode.com/xoai" xsi:schemaLocation="http://www.dspace.org/xmlns/dspace/dim http://www.dspace.org/schema/dim.xsd">
   <dim:field mdschema="dc" element="contributor" qualifier="author">Jin, Hanchen</dim:field>
   <dim:field mdschema="dc" element="contributor" qualifier="chair" lang="en_US">Zhang, Zhiru</dim:field>
   <dim:field mdschema="dc" element="contributor" qualifier="committeeMember" lang="en_US">Suh, Gookwon Edward</dim:field>
   <dim:field mdschema="dc" element="contributor" qualifier="committeeMember" lang="en_US">Albonesi, David</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="accessioned">2025-08-21T14:13:09Z</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="issued">2025-05</dim:field>
   <dim:field mdschema="dc" element="identifier" qualifier="other">ProQuest Submission ID: 14880</dim:field>
   <dim:field mdschema="dc" element="identifier" qualifier="other">ProQuest Publication ID: 31994595</dim:field>
   <dim:field mdschema="dc" element="identifier" qualifier="uri">https://hdl.handle.net/1813/117514</dim:field>
   <dim:field mdschema="dc" element="identifier" qualifier="doi">https://doi.org/10.7298/e8xg-2d42</dim:field>
   <dim:field mdschema="dc" element="identifier" qualifier="bibid">16938283</dim:field>
   <dim:field mdschema="dc" element="description" lang="en_US">182 pages</dim:field>
   <dim:field mdschema="dc" element="description" qualifier="abstract" lang="en_US">Sparse linear algebra (SLA) underpins a wide range of critical applications, including data analytics, graph processing, deep learning, and transformers. In light of the growing footprints of sparse processing and diminishing benefits of generalpurpose scaling, recent years have seen an active body of research on accelerating sparse kernels using specialized hardware. However, the diverse characteristics of inputs across different sparse applications– such as variations in dimensions, density, and non-zero value distributions– necessitate tailored architectural configurations and components, resulting in an extensive design space for exploration. To efficiently develop SLA accelerators, designers must rapidly explore the complex design and configuration space, making traditional, time-consuming simulation-based approaches unsuitable. This dissertation presents our recent research on analytical modeling for rapid design space exploration (DSE) and the efficient implementation of sparse accelerators. First, we introduce a novel and comprehensive analytical model capable of estimating the performance of various compute patterns for SLA kernels, factoring in the characteristics of input matrices and specific architecture configurations. Next, by leveraging the rapid DSE enabled by this analytical model, we develop Vesper, a versatile sparse accelerator equipped with High Bandwidth Memory (HBM) to efficiently handle datasets of varying dimensions and density. Vesper accelerates four key SLA kernels: sparse-matrix dense-vector multiplication (SpMV), sparse-matrix dense-matrix multiplication (SpMM), sparse-matrix sparse vector multiplication (SpMSpV), and sparse-matrix sparse-matrix multiplication (SpGEMM). Guided by the analytical model, Vesper is designed to be configurable and resource-efficient, supporting diverse compute patterns and kernels while achieving throughput comparable to or exceeding state-of-the-art sparse accelerators. Finally, we present GraphAxe, the first format-architecture co-designed hardware accelerator tailored for processing ultrascale (out-of-core) graphs using solid-state drives (SSDs). Specifically, we introduce a quantization-based doubly compressed sparse column (QDCSC) format to reduce SSD traffic. To further maximize bandwidth utilization, we implement SSD-aware mechanisms, including dynamic payload size adjustment and scoreboard-based coordination. Compared to the state-of-the-art CPU-based out-of-core graph computing system, GraphAxe delivers a 2.20x geometric mean runtime improvement and a 154.16x increase in energy efficiency.</dim:field>
   <dim:field mdschema="dc" element="description" qualifier="embargo">2028-06-18</dim:field>
   <dim:field mdschema="dc" element="language" qualifier="iso">en</dim:field>
   <dim:field mdschema="dc" element="rights" lang="*">Attribution 4.0 International</dim:field>
   <dim:field mdschema="dc" element="rights" qualifier="uri" lang="*">https://creativecommons.org/licenses/by/4.0/</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">Analytical modeling</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">Domain-specific architecture</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">Graph acceleration</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">Hardware accelerator</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">Sparse accelerator</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">Sparse linear algebra</dim:field>
   <dim:field mdschema="dc" element="title" lang="en_US">Analytical Modeling and Efficient Implementation for Specialized Hardware Acceleration of Sparse Applications</dim:field>
   <dim:field mdschema="dc" element="type" lang="en_US">dissertation or thesis</dim:field>
   <dim:field mdschema="dc" element="relation" qualifier="localuri">https://newcatalog.library.cornell.edu/catalog/16938283</dim:field>
   <dim:field mdschema="dc" element="format" qualifier="mimetype">application/pdf</dim:field>
   <dim:field mdschema="thesis" element="degree" qualifier="discipline">Electrical and Computer Engineering</dim:field>
   <dim:field mdschema="thesis" element="degree" qualifier="grantor">Cornell University</dim:field>
   <dim:field mdschema="thesis" element="degree" qualifier="level">Doctor of Philosophy</dim:field>
   <dim:field mdschema="thesis" element="degree" qualifier="name">Ph. D., Electrical and Computer Engineering</dim:field>
   <dim:field mdschema="dcterms" element="license">https://hdl.handle.net/1813/59810.2</dim:field>
   <dim:field mdschema="dspace" element="entity" qualifier="type">Publication</dim:field>
   <dim:field mdschema="cris" element="virtual" qualifier="collection" authority="https://cornell-ecommons.eks.prod.4science.cloud/handle/1813/47" confidence="600">Cornell Theses and Dissertations</dim:field>
   <dim:field mdschema="cris" element="virtual" qualifier="author">Jin, Hanchen</dim:field>
   <dim:field mdschema="cris" element="virtualsource" qualifier="collection">5893a6ea-7af3-41d7-abc6-04bcd26ab5df</dim:field>
   <dim:field mdschema="others" element="access-status">embargo</dim:field>
   <dim:field mdschema="others" element="access-status">embargo</dim:field>
   <dim:field mdschema="cerif" element="openaire" authority="" confidence="-1">&lt;Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="d1e2ccf9-68e8-4f75-9182-ed87e45b3fd1">
	&lt;Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843&lt;/Type>
	&lt;Language>en&lt;/Language>
   	&lt;Title>Analytical Modeling and Efficient Implementation for Specialized Hardware Acceleration of Sparse Applications&lt;/Title>
   	&lt;PublishedIn>
    	&lt;Publication>
      	&lt;/Publication>
   	&lt;/PublishedIn>
   	&lt;PublicationDate>2025-05&lt;/PublicationDate>
   	&lt;DOI>https://doi.org/10.7298/e8xg-2d42&lt;/DOI>
   	&lt;Authors>
      	&lt;Author>
        	&lt;DisplayName>Jin, Hanchen&lt;/DisplayName>
         	&lt;Affiliation>
         		&lt;OrgUnit>
         		&lt;/OrgUnit>
         	&lt;/Affiliation>
      	&lt;/Author>
	&lt;/Authors>
   	&lt;Editors>
	&lt;/Editors>
    &lt;Publishers>
        &lt;Publisher>
            &lt;OrgUnit />
        &lt;/Publisher>
    &lt;/Publishers>
    &lt;License>https://creativecommons.org/licenses/by/4.0/&lt;/License>
    &lt;Keyword>Analytical modeling&lt;/Keyword>
    &lt;Keyword>Domain-specific architecture&lt;/Keyword>
    &lt;Keyword>Graph acceleration&lt;/Keyword>
    &lt;Keyword>Hardware accelerator&lt;/Keyword>
    &lt;Keyword>Sparse accelerator&lt;/Keyword>
    &lt;Keyword>Sparse linear algebra&lt;/Keyword>
   	&lt;Abstract>Sparse linear algebra (SLA) underpins a wide range of critical applications, including data analytics, graph processing, deep learning, and transformers. In light of the growing footprints of sparse processing and diminishing benefits of generalpurpose scaling, recent years have seen an active body of research on accelerating sparse kernels using specialized hardware. However, the diverse characteristics of inputs across different sparse applications– such as variations in dimensions, density, and non-zero value distributions– necessitate tailored architectural configurations and components, resulting in an extensive design space for exploration. To efficiently develop SLA accelerators, designers must rapidly explore the complex design and configuration space, making traditional, time-consuming simulation-based approaches unsuitable. This dissertation presents our recent research on analytical modeling for rapid design space exploration (DSE) and the efficient implementation of sparse accelerators. First, we introduce a novel and comprehensive analytical model capable of estimating the performance of various compute patterns for SLA kernels, factoring in the characteristics of input matrices and specific architecture configurations. Next, by leveraging the rapid DSE enabled by this analytical model, we develop Vesper, a versatile sparse accelerator equipped with High Bandwidth Memory (HBM) to efficiently handle datasets of varying dimensions and density. Vesper accelerates four key SLA kernels: sparse-matrix dense-vector multiplication (SpMV), sparse-matrix dense-matrix multiplication (SpMM), sparse-matrix sparse vector multiplication (SpMSpV), and sparse-matrix sparse-matrix multiplication (SpGEMM). Guided by the analytical model, Vesper is designed to be configurable and resource-efficient, supporting diverse compute patterns and kernels while achieving throughput comparable to or exceeding state-of-the-art sparse accelerators. Finally, we present GraphAxe, the first format-architecture co-designed hardware accelerator tailored for processing ultrascale (out-of-core) graphs using solid-state drives (SSDs). Specifically, we introduce a quantization-based doubly compressed sparse column (QDCSC) format to reduce SSD traffic. To further maximize bandwidth utilization, we implement SSD-aware mechanisms, including dynamic payload size adjustment and scoreboard-based coordination. Compared to the state-of-the-art CPU-based out-of-core graph computing system, GraphAxe delivers a 2.20x geometric mean runtime improvement and a 154.16x increase in energy efficiency.&lt;/Abstract>
	&lt;Access xmlns="http://purl.org/coar/access_right" 
    >
    &lt;/Access>
&lt;/Publication>
</dim:field>
</dim:dim>
</metadata></record></GetRecord></OAI-PMH>