Storage technologies much faster than NAND flash aren’t expected to reach most smartphones and data centers for years, but preparations are already underway in order to make the most of them when they arrive.
Western Digital’s HGST subsidiary is demonstrating one advance in that effort this week, showing what it calls the world’s fastest SSD (solid-state drive) at the Flash Memory Summit in Santa Clara, California.
The device, which can be plugged into a server’s PCIe slot like any SSD, isn’t a new storage product but a platform for demonstrating a low-latency interface that the company developed with future solid-state media in mind. It implemented the experimental communications protocol in a Linux driver on the server and in the SSD’s embedded software.
The media at the heart of this SSD is PCM (phase change memory), an emerging technology that can make bits available to a host system many times more quickly than NAND flash can, according to Ulrich Hansen, HGST’s vice president of SSD product marketing. Where NAND takes about 70 microseconds to respond to a request for data, PCM can do so in about 1 microsecond, he said. However, it will take two to three years of hardware development before PCM gets dense and cheap enough to compete head-on with flash, according to Hansen.
Looking that far out, there are at least two other technologies in the running to become a faster and more durable alternative to flash. Hansen cited RRAM (resistive random-access memory) and MRAM (magnetic RAM). With low-latency technologies such as those and PCM, current interfaces between systems and memory won’t be able to take advantage of the new performance on offer, he said. Even NVMe, the interface that’s now emerging to standardize high-speed connections to flash, will be a bottleneck instead of an accelerator for those new generations of media.
“The reason we’ve chosen phase change for this particular ... card that we built was really just because it was available to us and it had the right attributes,” Hansen said. “We’re not really predicting which of the emerging memory technologies will ultimately go into production first.”
PCM stores data in a glass-like metal alloy called chalcogenide that can be changed from gas to crystalline form as its atoms are rearranged in response to a charge. The various states it can achieve—other, liquidy in-between states are also being experimented with—can be used to represent the ones and zeroes of binary data. Using more states lets the media pack more bits of data into the same space.
HGST’s interface runs on top of standard PCIe, which actually transports the data from flash to computer and will be fast enough for a long time, Hansen said. The higher-level interface defines things like how a command gets transferred and initiated and how the server knows a transmission is completed. HGST isn’t setting out to beat NVMe or to necessarily push its low-latency interface over any open standard.
“NVMe will evolve, right? So who knows?” Hansen said.
Also at Flash Memory Summit, HGST introduced its FlashMAX III PCIe accelerator, its third-generation server flash card for speeding up databases and other applications, and ServerCache, software for turning any SSD on a server into a cache. ServerCache uses technology derived from HGST’s Velobit acquisition for data reduction so SSD capacity can be used more efficiently, Hansen said. The FlashMAX III is set to ship in the current quarter, and ServerCache is available now for US$995 per physical server, with a free 30-day trial download.