The advent of FPGA-based flash storage cards enables data centers to customize their solution for maximum performance, storage capacity, and flash durability.
Social media, electronic commerce, and over-the-top consumption of video and audio content are radically changing the data center, which had been defined by its hardware components in the past. Today, the data center is rapidly becoming racks of commodity hardware components configured by software. One component benefiting from this change is solid-state storage in the form of flash memory. Internet giants such as Facebook, Microsoft, Baidu, Alibaba, LinkedIn, and others have opted to create their data centers from racks of commodity server blades tied to massive storage farms. Linking the two is high speed interconnect transporting continuous torrents of data that steadily grows over time. This is where flash becomes important.
To process this data more effectively, the storage hierarchy of the data center server and its associated disk farm has changed to include an additional layer of high-speed non-volatile memory (NVM) solid-state storage to mask the rotational latency of disk drives. According to market research firm IHS iSuppli, new enterprise products ranging from drives to caches to (disk) arrays have led to integration of solid state drives (SSD) into corporate storage architecture. IHS iSuppli says SSD shipments grew 50 percent in 2014 and predicts them to reach 189.6 million units in 2017 — close to half the size of the predicted hard disk drive (HDD) market of 397 million units that same year.
There are two areas where solid-state storage is making its presence known. One is where the solid-state storage cards plug directly onto the host server bus via a PCI Express interface. This is the hardware that enables split second E-commerce transactions, search, fraud detection, and other such time-critical operations. The other application is for block-based network storage linked to server farms via Fiber Channel SAN (storage area network), FCoE (Fiber Channel over Ethernet), and iSCSI (Internet Small Computer System Interface) arrays. These two applications provide servers with high-speed access while masking the latency inherent in rotating HDD memory.
Plug-in flash storage cards can provide up to six terabytes of high-speed cache per board for servers. These represent growth potential for flash storage card suppliers as the market is in its infancy and no one vendor has a dominant market position. Furthermore, flash storage cards themselves are undergoing change brought on by the newest generation of flash chips coming on the market. Since next-generation flash chips differ from one major chip vendor to another, flash storage card suppliers cannot build a single board solution that can easily mix these different flash chips. Furthermore, the total available number of flash storage cards is not sufficiently large to warrant an SoC (System on Chip) solution to the problem. This leaves a major opening for a flash storage card with an FPGA as its main computing element.
The flash chips becoming available to populate the next generation of flash storage cards are highlighted in this EETimes column from March 26, 2015, by Rick Merritt, which noted:
Micron said it will pack 256 Gbits into vertical NAND chips using two-bit per cell (a.k.a. MLC) technology and 384 Gbits in three-bit per cell (TLC) versions. By contrast, Samsung has been shipping since July its 850 series SSDs using 86-Gbit MLC and 128-Gbit TLC chips using vertical NAND. Both Samsung and Micron/Intel are fielding chips that stack 32 silicon levels.
At the same time, Toshiba began sampling the world’s first 48-layer three-dimensional (3D) stacked cell structure called the BiCS (Bit Cost Scalable) flash memory chip, that contains a two-bit-per-cell 3D-stacked cell structure flash affording 128Gb (16GB) of capacity that significantly reduces the overall size of the chip. Read More