Files in this item



application/pdfPATIL-THESIS-2016.pdf (6MB)
(no description provided)PDF


Title:Reliable spin-based computing systems
Author(s):Patil, Ameya D
Advisor(s):Shanbhag, Naresh R.
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):error resiliency
Abstract:Scaling of logic devices has enabled tremendous improvement in computational efficiency. However, computational scaling beyond the electronics based on Moore's law requires the adoption of alternate state variables including spin. Spin based devices offer several advantages such as low device count and non-volatility, and have the potential to beat the energy-delay product of CMOS. However, thermal noise in these devices makes their switching delay a random variable. Deterministic von Neumann style computing requires them to operate at worst case delay (and low error-rate), thereby completely offsetting the energy-delay benefits of spin devices and making them non-competitive against CMOS. In this thesis, we show that, by exploiting inherent device characteristics and architectural-level techniques, it is possible to shape the system-level output error distribution, thereby enabling effective error compensation and reliable system behavior. In particular, we demonstrate that, for a simple binary classifier, 33× improvement in accuracy over conventional design can be achieved while tolerating device error rate of 10%. This work paves a way towards the design of reliable spin-based systems using highly error prone, but energy-efficient spin devices.
Issue Date:2016-07-21
Rights Information:Copyright 2016 Ameya Patil
Date Available in IDEALS:2016-11-10
Date Deposited:2016-08

This item appears in the following Collection(s)

Item Statistics