Date of Award
7-8-2024
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
With the diminishing of Moore’s law and the end of Dennard scaling, alongside the explosion of data, the need for efficient computation—both in terms of performance and energy consumption—has become paramount. This is particularly crucial for processing large-scale data across various workloads. While a universal computation method is highly desirable, the inherent unique characteristics of different applications and datasets preclude a one-size-fits-all solution. It is usually the case that an appropriate computation technology choice could benefit a particular set of workloads and vice versa. Addressing this challenge requires designers to possess profound knowledge and insights into both hardware and software. Thus, the field of efficient computation involves many research questions driven by applications and platforms. In this dissertation, we focus on three representative types of applications: data integration, irregular graph processing, and graph neural networks. Our work in efficient computing leverages Near-Memory Processing (NMP) with 3D-stacked memory and High-Level Synthesis (HLS) on Field-Programmable Gate Arrays (FPGAs) to accelerate these workloads. We start the dissertation by adopting NMP technology based on 3D-stacked memory to accelerate data integration applications in terms of performance and energy consumption. We then present SuperCut, a novel hardware/software graph partitioning framework for near-memory graph processing. Subsequently, we describe GNNHLS, an open-source framework for the comprehensive evaluation of Graph Neural Network (GNN) kernels on FPGAs using high-level synthesis. Given the complexities of optimizing GNN HLS, we introduce HLPerf, an open-source, simulation-based performance evaluation framework for dataflow architectures that both supports early exploration of the design space and shortens the performance evaluation cycle.
Language
English (en)
Chair
Roger Chamberlain
Committee Members
Xuan Zhang