Storage

08:00 AM
Jim O'Reilly
Jim O'Reilly
Commentary
Connect Directly
LinkedIn
RSS
E-Mail
50%
50%

Will Phase Change Memory Upstage Flash?

HGST is touting a solid-state drive with a record 3 million IOPS using PCM. The technology looks to be a game changer, but it faces several challenges.

The nonvolatile memory space just got very interesting with the recent demonstration of a super-fast phase change memory (PCM) drive by HGST, a Western Digital company.

Unveiled at the Flash Memory Summit, the PCIe card is the first PCMunit to be demonstrated in product form. At 3 million IOPS using PCIe 2.0 connections, it's the fastest nonvolatile storage available. The technology also lays claim to ultra-low latency (around 1.5 microseconds) on short transfers at high speed.

These advances address major issues with flash memory performance, and once the product moves from the proof-of-concept stage to a full general release, it clearly could have quite an impact on the high end of the market. HGST's demo was neat. However, given the complexities of the new PCM technology and the fact that it was achieved with a new proprietary DCe interface that has yet to gain support in the industry, it may be quite a while before this memory approach hits the street.

DCe, HGST's approach to interfacing, looks like half of the still-evolving NVMe approach that the industry is bringing to market. Essentially, DCe is NVMe without the queues of completed I/Os and the protocol and transfers that go with them. It has fewer round-trip operations per I/O, and being lean allows more of the interface to carry payload.

HGST will have to build an industry consensus on DCe and get OS and hardware support. This is a double-edged sword, since a flash memory using DCe instead of NVMe could achieve roughly the same performance.

There are two important exceptions. One is read latency, where the PCM drive is almost 100 times better than the typical flash drive. That would be a tiebreaker for high-performance computing and some other applications. The other is write durability, which should not be an issue with PCM.

It's worth noting, however, that PCM write latencies are similar to flash.

HGST hasn't given any information on availability and price yet. Obviously, these will set boundaries for the PCM product in terms of market share.

The demo unit uses 45nm devices, so there is room for higher-density processes and thus faster devices to have an impact prior to release. The DCe interface appears to have a lot of merit as a solution for very fast storage, and it may gain ready acceptance, especially if HGST provides driver technologies.

So it seems that PCM drives have enormous potential to shake up the industry. At the same time, though, NAND flash is moving to a 3D packaging and device scheme. Companies like Micron and Intel are planning to stack die using through-silicon vias (TSVs) and take advantage of the Hybrid Memory Cube approach to memory interfacing to the CPU. This provides a more direct and much faster interface to memory than current DIMM approaches, with performance in the terabyte per second range for DRAM.

3D flash will also multiply capacities dramatically, with 16x or more boosts in areal density, and substantial power savings. For example, at the Flash Memory Summit, Samsung announced a triple-level cell (TLC) technology device with 32 layers of cells and a 2x density improvement per cell for a 64x capacity boost. The 3D approach will reduce cost, size, and power, and this will increase the barrier to entry for PCM.

The need to standardize DCe will delay things by a year or so, most likely, and it will open up room for further flash advancements. Meanwhile, ReRAM and any of the other promising flash replacements will be warming up for combat, and other players are working on PCM. My guess right now is that the end of 2015 will be a very interesting time in the nonvolatile space.

Jim O'Reilly was Vice President of Engineering at Germane Systems, where he created ruggedized servers and storage for the US submarine fleet. He has also held senior management positions at SGI/Rackable and Verari; was CEO at startups Scalant and CDS; headed operations at PC ... View Full Bio
Comment  | 
Print  | 
More Insights
Comments
Newest First  |  Oldest First  |  Threaded View
joreilly925
50%
50%
joreilly925,
User Rank: Apprentice
8/17/2014 | 12:01:28 PM
Re: niche market at first
I did some work on roadmaps and technology approaches for NVDIMM, which certainly meets the criteria of faster access and low latency. It's taken three years for the first couple of basic steps down the road, due to software issues and building a consensus.
soldack
50%
50%
soldack,
User Rank: Apprentice
8/17/2014 | 11:55:18 AM
niche market at first
If it is really much faster, it will command a premium in a niche market.  The interface and OS issues will have to be handled by their own drivers.  Others started down this path as well.  There is always an ultra high end user that is willing to pay for the very best and they are often the friends of new unproven technology.
joreilly925
50%
50%
joreilly925,
User Rank: Apprentice
8/14/2014 | 12:28:52 PM
Re: too many obstacles?
Marcia,

The Industry has just agreed on NVMe as a way to interface PCIe memory. DCe is in fact similar in that it uses a queue, but very different in how ti keeps track of completed operations.  

A good rule of thumb is to count the technologies that require a major evolution to get to market. In this case, it's PCM, device interfacing chips, DCe, and to do the full job, OS and application constructs to allow granular addressing. Hybrid Memory Cube also comes to mind.

My rule of thumb is that the pain-to-market (risk, engineering cost, tooling, etc)  is the square of that number. In this case its a score of 9 to 16. That's a big step and it won't happen overnight.
MarciaNWC
50%
50%
MarciaNWC,
User Rank: Strategist
8/14/2014 | 11:31:24 AM
too many obstacles?
"HGST will have to build an industry consensus on DCe and get OS and hardware support."

Altogther, it seems HGST faces too many challenges to really advance this technology in the market.
Cartoon
Slideshows
Audio Interviews
Archived Audio Interviews
Jeremy Schulman, founder of Schprockits, a network automation startup operating in stealth mode, joins us to explore whether networking professionals all need to learn programming in order to remain employed.
White Papers
Register for Network Computing Newsletters
Current Issue
Video
Twitter Feed