Design of High-performance, Resilient, STT-MRAM-based Main Memory

STT-MRAM is an exciting new memory technology that is being considered as a replacement candidate for SRAM in caches and DRAM in main memory. While the technology has adequate endurance and read/write latencies, challenges
to its adoption include relatively high write and read currents per cell and high error rates. We evaluate the feasibility of STT-MRAM as a main memory replacement technology for DRAM, analyze the challenges to its adoption and provide
solutions. We reduce power at a device-chip level by restricting the number of cells concurrently accessed within the chip for each request. We present a device architecture that enables this reduction without affecting its compatibility with existing
DRAM DIMM designs. For managing error rates we look in detail at the different error phenomena for STT -MRAM and examine the effectiveness of ECC technology and scrubbing to mitigate them. Using both analytical methods (for error modeling) and simulation methods (for power and performance quantification) we identify the STT-MRAM technology (characteristics) for which feasible ECC and scrubbing solutions can provide adequate error correction. Limiting the number of cells activated in a device for each access (for power feasibility) and adoption of error correction solutions have potential to reduce performance. We propose two optimizations to improve performance – command compaction and banklet architecture – and quantify their benefits with a suite of memory-intensive workloads.

By: Janani Mukundan, Alexandre Ferreira, Karthick Rajamani, Jente Kuang, Kyu-Hyoun Kim, Hillery Hunter, Luis Lastras, Xiaochen Guo

Published in: RC25679 in 2018

LIMITED DISTRIBUTION NOTICE:

This Research Report is available. This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g., payment of royalties). I have read and understand this notice and am a member of the scientific community outside or inside of IBM seeking a single copy only.

RC25679.pdf

Questions about this service can be mailed to reports@us.ibm.com .