Abstract

Emerging hybrid-CMOS nanoscale devices and architectures offer greater degree of integration and performance capabilities. However, the high power densities, hard error frequency, process variations, and device wearout affect the overall system reliability. Reactive design techniques, such as redundancy, account for component failures by mitigating them to prevent system failures. These techniques incur high area and power overhead. This research focuses on exploring hybrid CMOS/Resistive RAM (RRAM) architectures that enhance the system reliability by performing computation in RRAM cache whenever CMOS logic units fail, essentially masking the area overhead of redundant logic when not in use. The proposed designs are validated using the Gem5 performance simulator and McPAT power simulator running single-core SPEC2006 benchmarks and multi-core PARSEC benchmarks. The simulation results are used to evaluate the efficacy of reliability enhancement techniques using RRAM. The average runtime when using RRAM for functional unit replacement was between ~1.5 and ~2.5 times longer than the baseline for a single-core architecture, ~1.25 and ~2 times longer for an 8-core architecture, and ~1.2 and ~1.5 times longer for a 16-core architecture. Average energy consumption when using RRAM for functional unit replacement was between ~2 and ~5 times more than the baseline for a single-core architecture, and ~1.25 and ~2.75 times more for multi-core architectures. The performance degradation and energy consumption increase is justified by the prevention of system failure and enhanced reliability. Overall, the proposed architecture shows promise for use in multi-core systems. Average performance degradation decreases as more cores are used due to more total functional units being available, preventing a slow RRAM functional unit from becoming a bottleneck.

Library of Congress Subject Headings

Nonvolatile random access memory; Memristors; Metal oxide semiconductors, Complementary--Design and construction; Nanoelectronics

Publication Date

8-1-2012

Document Type

Thesis

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Kudithipudi, Dhireesha

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TK7895.M4 C38 2012

Campus

RIT – Main Campus

Share

COinS