Files in this item

FilesDescriptionFormat

application/pdf

application/pdfWANG-THESIS-2020.pdf (767kB)Restricted to U of Illinois
(no description provided)PDF

Description

Title:Memory-centric approximate computing
Author(s):Wang, Dong Kai
Advisor(s):Kim, Nam Sung
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:M.S.
Genre:Thesis
Subject(s):Approximate computing
Computer architecture
Abstract:As Moore's law continues to decline, diminishing benefits of transistor scaling necessitate a move towards specialized hardware. Approximate computing is one type of specialization that has shown promise in improving the efficiency of general-purpose processors. Fortunately, with increasing demand for data collection and processing across industry, a wide range of modern applications operate on real-world data with properties suitable for approximation. To exploit data patterns and repetitions in these applications, we propose Approximate Algebraic Memory (A2M), a specialized memory model that uses finite degree polynomials to approximate discrete ranges of memory data. A2M uses dedicated hardware to derive and store polynomial coefficients rather than memory data. In error resilient workloads, A2M can effectively reduce memory size and enable direct computation on memory content. We evaluate an on-chip implementation of A2M for general-purpose processors. Experiment results show that for CPU workloads, A2M yields minimal error (< 1%) at a fixed compression ratio of 16, and improves performance by 11.3% on average.
Issue Date:2020-05-12
Type:Thesis
URI:http://hdl.handle.net/2142/108178
Rights Information:Copyright 2020 Dong Kai Wang
Date Available in IDEALS:2020-08-26
Date Deposited:2020-05


This item appears in the following Collection(s)

Item Statistics