No toys to throw from the PRAM
Phase-change memory could be going nowhere slowly
Comment Phase-change memory (PCM or Phase-change RAM - PRAM) seems to be changing its phase, from promising-newcomer-technology to fading-candidate-going-nowhere.
PCM is a memory technology involving a change of material state and electrical resistance in a memory cell's chalcogenide layer. The theoretical attractions are that it is non-volatile, like flash, bit-addressable, like DRAM, and high-density. In practice it has proved difficult to commercialise. Problems with the technology have taken years to solve, resulting in low density product using relatively large process geometries, such as 90nm and 65nm – and apparently large power draws as well.
Meanwhile NAND flash memories are currently being fabricated using sub-30nm processes with sub-10nm geometries in prospect.
There are two PCM developers: Micron, through acquiring Numonyx, and Samsung, and they have to contend with other follow-on technology candidates to succeed NAND, such as Resistive RAM and HP's Memristor, while developing their PRAM products. It could be that NAND is viable for long enough for one of the PRAM alternatives to come to the fore and render PRAM pointless.
Our understanding is that Samsung has shipped limited quantities of a 512Mbit PRAM device but neither it nor Micron has yet shipped a 1Gbit capacity chip. The Micron Numonyx 1Gbit device, which was supposed to ship last year, uses a 45nm process whereas Samsung's is based on a 58nm process.
Talking at CeBIT, Steve MacDonald, an EMEA executive at Micron, one of the two PCM suppliers, said Micron is continuing its PCM development work and: "Eventually it may replace NOR for entry-level NOR applications. It is already in the embedded market with serial and parallel PCM ... We believe NAND will be viable for the next five years [with] geometries below 20nm. But below 10nm? I don't know."
He said that: "Samsung has announced PCM but we haven't seen anything coming out yet."
Samsung talked about its PRAM toys at the 2011 International Solid-State Circuits Conference in San Francisco last month, and said OEMs face a non-compatibility obstacle in moving from flash memory to PRAM. But it seems to El Reg the OEMs also face a bigger obstacle in that PRAM doesn't offer them anything significantly better than their existing flash memory components and may not do so until 2017.
Technologies like that in Anobit's digital signal processing controllers, and over-provisioning to combat limited endurance, as well as better flash-optimised system software, could enable NAND to persist to the point that PRAM becomes redundant.
With millions of dollars invested in the technology it will be a hard decision for either Micron or Samsung to move on from PRAM to something else. Perhaps the pair of them would be better off partnering in PRAM and not competing? ®
COMMENTS
Ok I'll bite...
(or should that be Byte in this context ;)
To be fair new technology always takes a while to displace the incumbent technology. For example no moving parts Flash is better than spinning disks, yet we still have cheap disks which get ever better. Its going to take some time before Flash reaches the densities needed to displace spinning disk drives in all market areas.
The same is true of even newer technology, like PCM, where its first sold to high value niche markets, before they can then start to sell it to higher volume mass markets, where its price will finally fall.
That's not to say PCM is going to finally win, there are as you say other competitors fighting for the same place in the market. PCM is much faster than Flash so it will win in any market that needs speed. PCM also suffers less from Memory Wear (PCM survives for about ten thousand times more writes than Flash so there will definitely be markets for PCM over Flash).
I think the biggest differentiators from Flash will be memory Density and Memory Wear (as they are the mass markets), but which memory technology will win in each of these areas, who knows yet. But bigger densities will win market share in all existing storage applications. But the even bigger one (and the biggest unknown) is going to be Memory Wear. If someone comes up with a technology that doesn't suffer from Memory Wear but also has the large storage capacity of Flash combined with the speed of DRAM, then they win massively if that is comparably priced.
The biggest problem I think PCM has is simply the potential Memristors have. That will (currently) scare off investors in PCM and other technology until weaknesses in Memristors are more clearly confirmed as weaknesses (which market areas will it be slow to move into highlighting its weaknesses). Currently its hailed (often by its creators) as the holy grail of circuit design and in many ways it is *in theory* ... The problem is we have yet to see mass market Memristor drives. Memory Wear however is likely to be Memristor's Achilles Heel. Its not going to be a DRAM replacement until the Memory Wear problem is totally solved and as far as I know it hasn't been (it would be huge news if it has been solved!).
So anyway PCM isn't a direct competitor to Flash, they have different markets, its just the elephant in the room is Memristors which would worry any investors in other memory technologies like PCM, at least currently.
Let me correct that for you.
"To be fair new technology always takes a while to displace the incumbent technology. For example no moving parts Flash is better than spinning disks, yet we still have cheap disks which get ever better. Its going to take some time before Flash reaches the densities needed to displace spinning disk drives in all market areas."
No, flash is better than spinning disks FOR SOME APPLICATIONS. For other applications, due to densities, spinning rust is better. For even other applications, a mixture of the two is ideal. When/If flash achieves the densities of spinning rust, it will likely be better, as long as there isn't a trade off to achieve those densities, but until then it is not better.
I Just Reread The...
...HP memristor article, and it flashed* on me that memristors could be easily mask programmed, yet electrically reprogrammed.
No idea how much money is spent burning firmware images into memory for mass market devices, but if it's not patented yet, remember you heard it here first.
*intentional inappropriate use of industry jargon in stoner context
