Although NAND flash might thrive in enterprise solid-state storage through the end of the decade, potential successors are already starting to generate attention. These include phase-change memory (PCM), magnetoresistive RAM (MRAM) and resistive RAM (RRAM).
The past couple of years have brought breakthroughs in these new memory technologies that could lead to boosts in performance and endurance far beyond the current state of the art.
IBM Research, for instance, turned heads over the summer with its announcement that thermal-based PCM might one day enable systems to retrieve data 100 times faster than NAND flash and to endure at least 10 million write cycles, an exponential improvement over the 30,000 write cycles of enterprise multi-level cell (eMLC) flash.
Until an alternative technology meets or beats NAND on cost, it’s unlikely to become a serious contender in enterprise storage, no matter how great its technological advantage.
IBM’s Zurich-based research lab claimed to achieve the reliable storage of multiple bits per cell after monitoring the resistance of the PCM material for six months to confirm the cell’s stability. Haris Pozidis, manager of memory and probe technologies at IBM Research in Zurich, said multi-level cell (MLC) PCM could make its way into enterprise servers and storage by 2016. He said the technology could benefit applications such as “big data” analytics and cloud computing.
That prediction comes with a caveat. Pozidis said the enterprise timeline hinges on mobile phone and memory manufacturers taking up MLC-PCM as a NOR flash replacement in the next couple of years.
Because IBM doesn’t make memory devices or solid-state drives (SSDs), the company relies on external manufacturers to license and produce its MLC-PCM. Pozidis said IBM expects that to happen, but he conceded that it’s no “sure thing.”
PCM and "racetrack memory"
PCM is only one of the memory technologies IBM has been promoting. The company’s Almaden, Calif.-based research center has been touting a magnetic technology called “racetrack memory,” which works by sliding magnetic bits back and forth along nanowire “racetracks.”
In an IBM-authored “Icon of Progress” story on its website, the company claimed that racetrack memory, with a three-dimensional microchip, has “the potential to replace nearly all forms of conventional data storage.” The technology could enable mobile phones, laptop computers and business-class servers to store more information by a factor of 100 times and provide access at far higher speeds, according to IBM.
Stuart Parkin, a fellow and manager of the Magnetoelectronics Group at IBM Research in Almaden, said the technology offers additional advantages over flash, including no wear-out or write limitations, and has the potential to beat NAND on cost. He predicted the nascent technology could emerge in products in the next five to seven years, potentially placing it on a similar timeline with PCM.
IBM's Pozidis, however, said PCM is “the nearer-term memory technology” and “much more advanced” than IBM racetrack memory.
While IBM licenses more than a billion dollars a year of its intellectual property, and the company’s work on new memory technologies shows promise, its PCM and racetrack efforts illustrate the uncertainty over which memory technology ultimately might usurp NAND flash.
“I don’t really take what IBM says too seriously,” said Greg Wong, founder and principal analyst at Forward Insights in North York, Ontario. “If their business was based on making that memory and making it work and commercializing it, that’s a different story. They don’t really have skin in the game.”
PCMS, STT-MRAM and RRAM
Major semiconductor manufacturers that do have "skin in the game" are exploring multiple technologies, including PCM, MRAM and RRAM. But until an alternative technology meets or beats NAND on cost, it’s unlikely to become a serious contender in enterprise storage, no matter how great its technological advantage.
“The next closest thing is probably still five times more expensive,” said Kevin Kilbuck, director of marketing in the NAND solutions group at Micron Technology Inc., commenting on the company’s studies of the alternatives.
In the meantime, Micron works in many areas, including PCM (through its 2010 acquisition of Numonyx B.V.) and Spin-Transfer Torque Magnetoresistive RAM (STT-MRAM). Kilbuck said some technologies may wind up complementary to NAND flash.
“They still might have merit on their own,” Kilbuck said. “It’s too early to say there's a single successor.”
Troy Winslow, the director of product marketing in Intel Corp.’s Non-Volatile Memory Solutions Group, said via email that Intel thinks a stacked PCM variant known as phase change memory and switch (PCMS), for its PCM element layered with an Ovonic Threshold Switch, has greater potential than MRAM in enterprise systems.
Intel and Numonyx announced in 2009 that they had demonstrated that a 64 Mb test chip could stack multiple layers of PCM arrays within a single die and pave the way for memory devices of greater capacity, performance and scalability, as well as lower power consumption.
But, Winslow added, “NAND still has a long life in enterprise solutions. Even when a new technology emerges in the next few years, the transition will take place over several years.”
In July 2011, Hynix Semiconductor Inc. and Toshiba Corp. issued a joint press release about their agreement to collaborate on STT-MRAM, a technology they said would be a good fit for smart phones, with its advantages of high speed, high capacity and low power consumption. Toshiba declined further comment, but others claim STT-MRAM eventually might find a home in enterprise systems, too.
A month later, Samsung Electronics Co. Ltd. announced its acquisition of Grandis Inc., noted for its work with STT-MRAM. Samsung will merge Grandis into its R&D operations, but STT-MRAM represents only one of many NAND alternatives in which the company is investing.
“Samsung, the world’s largest NAND shipper, is not going to let NAND die without trying everything in the book,” said Ray Lucchesi, president and founder of Silverton Consulting Inc. in Broomfield, Colo. “There’s a lot of money and a lot of companies behind NAND. It’s not going to go away any time soon.”
Forward Insights' Wong doesn’t foresee MRAM or even the stackable variant of PCM as being able to compete with NAND on price in the long run. He said manufacturers are currently looking at PCM and MRAM to supplement or replace parts of DRAM, serving essentially as a type of non-volatile RAM that can retain data even when the power is off, unlike DRAM. He said the technologies could be most useful with random write-intensive workloads that have high I/O requirements, such as databases and financial applications.
But, rather than PCM or MRAM, RRAM might hold the greatest potential to replace NAND, according to Wong. He said that all the big memory companies are working on RRAM, which operates by changing the resistance between two electrical terminals. Hewlett-Packard (HP) Co. also has a version that it calls memristor.
“RRAM theoretically is a scalable technology,” Wong said, “but what you need for any technology to replace NAND is it has to be stackable. That’s a fundamental challenge.”
NAND Flash successors affected by tech's extended lifespan
In the meantime, predicting which promising up-and-coming technology will succeed NAND flash has become especially difficult because NAND flash is enjoying a far longer lifespan than anyone expected when the technology first emerged.
“These absolute geniuses in the R&D labs keep coming up with ways to push it just one or two more process steps, and they’ve been doing that for almost 10 years,” said Jim Handy, founder and chief analyst at Object Analysis in Los Gatos, Calif. “Who’s to say they’re not going to do it for another 10 years?”
Handy recalls a 2003 Intel Developer Forum presentation, when an executive asserted that NAND flash wouldn't scale beyond a 60 nanometer (nm) process geometry. Partners Intel and Micron subsequently scaled the technology from 50 nm to 34 nm and 25 nm, and now to 20 nm. Handy expects flash will ultimately shrink to 10 nm and run out of steam at about 8 nm.
With the scaling comes a significant drop in cost. Packing more into the flash cells, with two-bit-per-cell MLC or the emerging consumer-centric triple-level cell (TLC) flash also lowers the price. But the innovation carries the unfortunate consequence of worsening endurance, performance and reliability.
So far, manufacturers have been able to make the two-bit MLC NAND flash usable in enterprise data storage with resourceful fixes such as sophisticated controller technology, error correction code (ECC) and wear leveling. However, they may be running out of tricks with the existing floating-gate technology.
One promising development to extend the life of flash is 3D NAND, which stacks the memory cells in three dimensions. Micron Technology’s Kilbuck likened it to a multistory office building vs. planar NAND flash’s single-story building spread out over a flat, wide campus.
“There’s still a lot of debate in the industry on when the current planar NAND cell is going to reach its scaling limits,” Kilbuck said. “I’ve heard people say it could be the high teens. I’ve heard other people say NAND is going to scale down into the single digits before we have to go to something like 3D NAND.”
Kilbuck said the industry won’t know until it starts manufacturing 3D NAND at high volume what sort of challenges lie ahead and how long the 3D technology might be able to extend NAND’s lifespan.
“The end of flash scaling is what everybody’s worried about and why they’re placing bets on other technologies,” Object Analysis' Handy said. But he thinks it’s too early to tell whether PCM, MRAM, RRAM or some other technology will win. “If it takes another 10 years before NAND flash can’t be shrunk any more, then that’s 10 more years for some dark horse to come out of a corner somewhere and become the leading candidate.”
This was first published in November 2011