Error characterization and correction techniques for reliable STT-RAM designs.

摘要

The concerns on the continuous scaling of mainstream memory technologies have motivated tremendous investment to emerging memories. Being a promising candidate, spin-transfer torque random access memory (STT-RAM) offers nanosecond access time comparable to SRAM, high integration density close to DRAM, non-volatility as Flash memory, and good scalability. It is well positioned as the replacement of SRAM and DRAM for on-chip cache and main memory applications. However, reliability issue continues being one of the major challenges in STT-RAM memory designs due to the process variations and unique thermal uctuations, i.e., the stochastic resistance switching property of magnetic devices.;In this dissertation, I decoupled the reliability issues as following three-folds: First, the characterization of STT-RAM operation errors often require expensive Monte-Carlo runs with hybrid magnetic-CMOS simulation steps, making it impracticable for architects and system designs; Second, the state of the art does not have suffciently understanding on the unique reliability issue of STT-RAM, and conventional error correction codes (ECCs) cannot effciently handle such errors; Third, while the information density of STT-RAM can be boosted by multi-level cell (MLC) design, the more prominent reliability concerns and the complicated access mechanism greatly limit its applications in memory subsystems.;Thus, I present a novel through solution set to both characterize and tackle the above reliability challenges in STT-RAM designs. In the first part of the dissertation, I introduce a new characterization method that can accurately and effciently capture the multi-variable design metrics of STT-RAM cells; Second, a novel ECC scheme, namely, content-dependent ECC (CD-ECC), is developed to combat the characterized asymmetric errors of STT-RAM at 0→1 and 1→0 bit flipping's; Third, I present a circuit-architecture design, namely state-restricted multi-level cell (SR-MLC) STT-RAM design, which simultaneously achieves high information density, good storage reliability and fast write speed, making MLC STT-RAM accessible for system designers under current technology node. Finally, I conclude that effcient robust (or ECC) designs for STT-RAM require a deep holistic understanding on three different levels{device, circuit and architecture. Innovative ECC schemes and their architectural applications, still deserve serious research and investigation in the near future.