Dissertation/Thesis Abstract

Design and Investigation of Genetic Algorithmic and Reinforcement Learning Approaches to Wire Crossing Reductions for pNML Devices
by Gunter, Alexander, M.S., The University of Mississippi, 2018, 110; 13419369
Abstract (Summary)

Perpendicular nanomagnet logic (pNML) is an emerging post-CMOS technology which encodes binary data in the polarization of single-domain nanomagnets and performs operations via fringing field interactions. Currently, there is no complete top-down workflow for pNML. Researchers must instead simultaneously handle place-and-route, timing, and logic minimization by hand. These tasks include multiple NP-Hard subproblems, and the lack of automated tools for solving them for pNML precludes the design of large-scale pNML circuits.

In this thesis we investigate potential solutions to the problem of wire crossing reduction in pNML circuits. Although pNML permits 3D architectures, planar designs are still preferred for the ease of fabrication; and reducing out-of-plane nanomagnets reduces risks of fabrication effects. We have found no existing work on this problem in pNML, and existing work for related technologies does not consider variations of the wire crossing problem that are specific to pNML. We present and evaluate two algorithms designed to address this research gap. The first is a genetic algorithm utilizing a multi-chromosome encoding of graph embedding, and the second is a deep reinforcement learning algorithm utilizing a similar encoding and Q-Learning. We also present a naïve NP-time randomized search algorithm for use as a baseline. The presented reinforcement learning algorithm proved unacceptably slow and ineffective, but our genetic algorithm removed significantly more wire crossings than the random search did on the ISCAS ’85 combinational benchmarks.

Indexing (document details)
Advisor: Morrison, Matthew
Commitee: Chen, Yixin, Gordon, Richard
School: The University of Mississippi
Department: Electrical Engineering
School Location: United States -- Mississippi
Source: MAI 58/04M(E), Masters Abstracts International
Source Type: DISSERTATION
Subjects: Computer Engineering, Computer science
Keywords: Crossing reduction, Electronic design automation, Genetic algorithm, Nanomagnet logic, Quantum-dot cellular automata, Reinforcement learning
Publication Number: 13419369
ISBN: 9781392002735
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest