Flash storage first appeared in datacentre products only six years ago, and its enterprise usage is still far from saturation point.
But semiconductor manufacturers are developing other solid-state technologies, some of which could succeed flash and other forms of solid-state memory in the not-too-distant future.
Why is this and when is this going to happen?
Flash up against the wall?
Chip makers have been approaching the limits to which they can reduce NAND flash manufacturing process sizes for some time.
This has led to speculation that the downward curve of flash prices seen over the last several years would flatten out. And then the forecast was that flash would hit a price wall over which it would be unable to climb.
However, in 2013, the time span for that forecast was effectively delayed for five or six years by Samsung’s announcement it had begun volume production of NAND flash chips featuring a 3D internal architecture.
This architecture superseded the 2D or planar architecture used in Samsung's previous NAND flash chips, and it restored the potential for further reductions in manufacturing process size.
Other manufacturers are set to follow Samsung's lead into 3D flash, meaning that all suppliers will continue to see reduced manufacturing costs.
Flash prices are now widely expected to continue to descend at about the same rate as they have been for the last few years.
This will continue until the process size for 3D NAND flash has reached about 10nm at around the end of this decade. At that point, the consensus is there will be little or no prospect of reducing NAND flash costs further.
Read more about flash storage
- Big storage takes two routes in flash array shakedown
- Flash array market roundup: The startups
- Big storage turns the tide in the hybrid flash array market
- PCIe SSD flash storage roundup: The few become fewer
- Flash caching software market roundup
- Podcast: MLC, eMLC, SLC, TLC – what they are and what they’re good for
- Smart and Diablo launch Ulltra DIMM memory channel storage
- Atlantis VDI storage helps hospitals save £500,000 a year
- University does the maths and dumps HDS for Nexenta storage software
- What’s wrong with flash storage: And what will come after?
Candidates to replace flash storage
Resistive RAM or ReRAM – also known as RRAM and memristor memory (pictured) – stores data by flipping resistors between two stable states. Compared to NAND flash, it promises higher performance, vastly longer life and lower power consumption.
ReRAM is being developed by a number of suppliers, of which HP has the highest profile. HP describes memristor as a form of ReRAM and says it can be used to create solid-state storage that provides 10 times the performance of NAND flash, with 10 times less power consumption and longer life.
However, the ReRAM that HP envisages will not be fast enough to compete with SRAM or DRAM and, instead, will only be a challenger to flash. This raises the possibility that HP's memristor technology may not reach the market until NAND flash technology – and the heavy investments already made in NAND manufacturing capacity – have been fully played out.
HP has not yet brought its memristors to market. In 2010, the IT giant announced a partnership with Korean memory and flash chip maker SK Hynix, in which HP would provide the semiconductor design, and Hynix would provide the manufacturing expertise needed for volume manufacture of memristor-based ReRAM.
Since then, HP has made periodic promises about the future impact of the memristor, but the company has steered clear of making public statements about exactly when volume production of the devices will begin.
Other investors in ReRAM include Panasonic, Toshiba and Micron.
Magneto-resistive random-access memory (MRAM) is already being used as an alternative to SRAM and offers similar performance levels but is more physically dense.
But MRAM is currently set to be more expensive than both DRAM and SRAM, and this means, initially at least, that its biggest appeal will be in applications where its non-volatility will compensate for its higher cost.
However, a variant of MRAM that has recently entered volume production may enable much lower costs and if that happens, MRAM could also be a rival to NAND and NOR flash, against which it offers a far longer life and greater performance.
Phase-change memory (PCM) is also known as PRAM and PCRAM. The technology has the potential to provide non-volatile storage that heavily outperforms NAND flash and offers a considerably longer write life.
IBM has described PCM as having “superb” scalability due to its ability to be produced at small process sizes with consequently higher data densities and lower costs.
The potential uses of PCM range from a replacement to NAND flash through an intermediate storage medium sitting between flash and DRAM in terms of performance and cost, to the ultimate role as a universal or “storage class” memory that would replace the entire portfolio of SRAM, DRAM, NOR flash and NAND flash.
Suppliers including Micron, IBM, Intel and Samsung have been investigating PCM for several years. In 2012, IBM announced a joint development deal with giant flash chip maker SK Hynix, under which Hynix would manufacture the jointly developed PCM chips. However, IBM has yet to release PCM products in volume.
Which flash successor will win out?
We make no predictions about which technology will most likely succeed flash. All three candidates have been in development for several years, and there have already been false starts around volume manufacture.
If any of them do eventually replace flash, we don’t believe this would happen until well into the next decade.
Flash has huge momentum because of the massive investments already made in flash manufacturing plants. And there would almost certainly be a transitional period, during which production volumes of the new storage medium would have to ramp up for its prices to fall and become competitive with flash.
There is also the question of which market will generate the largest demand for a flash successor. The answer is likely to be the consumer market for handheld and mobile devices – as it was for the NAND flash now used in enterprise storage. That fact will also influence developments.
Tim Stammers is senior analyst and Simon Robinson research vice president at 451 Research.
This was first published in April 2014