Advanced search
Start date
Betweenand


Analysis of wear-out of error correction techniques in phase-change memories

Full text
Author(s):
Caio Hoffman
Total Authors: 1
Document type: Master's Dissertation
Press: Campinas, SP.
Institution: Universidade Estadual de Campinas (UNICAMP). Instituto de Computação
Defense date:
Examining board members:
Guido Costa Souza de Araújo; André Rauber Du Bois; Mario Lúcio Cortês
Advisor: Rodolfo Jardim de Azevedo; Guido Costa Souza de Araújo
Abstract

Phase-change memory brings new opportunities for the electronics industry. Due to projections of high scalability of the fabrication process, PCM is seen as a new main memory in computing systems, replacing the traditional DRAM, whose scale problems require new future technologies that are still unknown. However, PCM has low endurance when compared with DRAM and robust failure recovery techniques are required to increase its lifetime. To address that, some error correcting techniques have been proposed, based on the non-volatile features of the PCM memories. In this work, we model and analyze the bit-ip probabilities of five such techniques (ECP, parity, SECDED, SAFER and BCH), in order to evaluate its impact to the wear out of the PCM. Using the bit-ip rate of 15%, obtained experimentally from the execution of the SPEC2006 benchmark, we mathematically modeled and simulated these techniques using both an empirical and theoretical probability rates. Our results show a clear degradation in techniques that use error-correcting codes, contradicting the previous results in the literature. Only ECP has not shown any degradation. We have also done power analyses of the above listed techniques so as to relate the endurance and the energy required by each technique. Again, the ECP stood out in the results, like SAFER as well. Finally, analytical probabilistic models for ECP and SECDED were proposed and an analysis of PAYG technique (based on ECP's analytical model) was performed (AU)

FAPESP's process: 11/05266-3 - Wear-leveling techniques for PRAM memories
Grantee:Caio Hoffman
Support Opportunities: Scholarships in Brazil - Master