Nonvolatile memory outlook: Technology driven or application driven

The last decade has seen significant progress in emerging nonvolatile memory (NVM) technologies including Spin Torque Transfer RAM (STT RAM), phase change memory (PCM) and resistive RAM (RRAM). Until now, the key industry players have all demonstrated Gb-scale capacity in advanced technology nodes, including 1Gb PCM at 45nm by Micron [1], 8Gb PCM at 20nm by Samsung [2], 32Gb RRAM at 24nm by Toshiba/Sandisk [3], 16Gb conductive bridge (CBRAM) at 27nm by Micron/Sony [4], 128Gb 3D XPoint technology by Micron/Intel [5] and 1Gb STT-RAM with DDR4-compatible interface by Everspin [6]. Even with successful commercialization, the insertion of these technologies to exiting computer system as a direct drop-in replacement seems not being effective. The fundamental reasons for that are 1) Technically, the inherent nature of these technologies does not align well with either main memory or persistent storage in terms of cost-per-bit, latency, power, endurance, and retention. 2) Economically, besides more investment to existing memory manufacturing facilities for producing these new technologies, it is difficult to convince end-users to switch to a new technology as long as they can still use DRAM or Flash for the same purpose, unless significant benefits are provided.