Trends in Storage Technologies E. Eleftheriou, R. Haas, J. Jelitto, M. Lantz and H. Pozidis IBM Research – Zurich, 8803 R ¨ uschlikon, Switzerland Abstract The high random and sequential I/O requirements of contemplated workloads could serve as impetus to move faster towards storage-class memory (SCM), a technology that will blur the distinction between memory and storage. On the other hand, the volume of digital data being produced is increasing at an ever accelerating pace. Recent technology demonstrations indicate that low-cost, low-power tape storage is well positioned to scale in the future and thus serve as archival medium. In this paper, we will look at the implications of these technology extremes, namely, SCM and tape, on the storage hierarchy. 1 Introduction One of the challenges in enterprise storage and server systems is the rapidly widening gap between the per- formance of the hard-disk drives (HDD) and that of the remainder of the system. Moreover, trying to narrow this gap by increasing the number of disk spindles has a major impact on the energy consumption, space usage, and cost. Given that major improvements in HDD latency are not expected, research and development efforts have shifted towards semiconductor memory technologies that not only complement the existing memory and storage hierarchy but also reduce the distinction between memory (fast, expensive, volatile) and storage (slow, inexpensive, nonvolatile). In the past, Flash memory technology has been driven by the consumer market alone. The current cost per GB and latency characteristics of NAND Flash make it an interesting candidate to bridge the widening per- formance gap in enterprise storage systems. In particular, the introduction of multi-level cells (MLC) further decreased the cost as compared to single-level cells (SLC), albeit at the expense of reliability and performance. The absence of moving parts in NAND Flash enhances the ruggedness and latency, and reduces the power consumption, but its operation and structure pose specific issues in terms of reliability and management over- head, which can be addressed in the hardware or software of a Flash controller. Although NAND Flash has already established itself in the memory and storage hierarchy between DRAM and HDDs, it still cannot serve as a universal memory/storage technology. In other words, it does not fully qualify as a storage-class memory (SCM) technology [1]. There are various technologies that could eventually qualify as SCM which combine the performance benefits of solid-state memories with the low cost and the large permanent storage capabilities of magnetic storage. These include ferroelectric, magnetic, phase-change, and resistive random-access memories, including perovskites and solid electrolytes, as well as organic and polymeric memories [2]. It is widely believed that the prime candidate of all these technologies to become SCM is phase-change memory (PCM). Copyright 2010 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE. Bulletin of the IEEE Computer Society Technical Committee on Data Engineering 1