Slashdot Mirror
IBM Creates Multi-Bit Phase Change Memory
Lucas123 writes "In what is likely to be a strong rival to NAND flash memory, IBM today announced it has been able to successfully store more than one bit of data per cell in a more stable non-volatile memory called phase-change memory (PCM). Unlike NAND, Previously, PCM couldn't contend with flash because of its low capacity points. PCM does not require that data be erased before new data is written to it, which reduces write amplification or wear out and it has 100 times the write performance of flash. IBM researchers say they plan to license the technology to memory manufacturers instead of producing it themselves."
I've heard about so many sea-change advances in technology these days and none of them pan out. Please excuse me while I go invest all of my fortune in IBM stock.
PCM. Not to be confused with Pulse Code Modulation.
Dad, tell us again about how you used to store your data on spinning disks....
I this story went up as I hit submit on it.
I do not have a sig. You are hallucinating.
we all know researchers don't take business decisions
they plan to license the technology to memory manufacturers instead of producing it themselves
This is code for 'send it to Asia.' HP is doing the same thing with their memristor based ReRAM, licensing it to Hynix. Guess we're all done building chip foundries anywhere in the West. As goes your manufacturing base, so goes your technology, just like we we're told would happen.
I like flash, in fact I just bought some SSDs, but it does have some problems that really need addressing long term. This stuff looks like maybe it'll be the solution. Also it looks pretty workable, it isn't pie in the sky. According to the article, PCM already exists and is in use. This is just an improvement on it. They also claim widespread use by 2016 of the new tech.
So here's hoping, this looks like it may be what we need to really kick off the move to solid state storage.
...much more efficient swap space! :) This sounds great.
I know it doesn't need to be erased in bulk, but does it need to be written in bulk, or does it have limited read or write cycles?
It's density+cost per bit that will change things. PCM is still orders of magnitude slower than SRAM and DRAM and the Memristor (HP) may still beat it to market for the aforementioned parameters. Nice that so many irons are in the fire to retire spinning rust, can't happen soon enough for me.
Organization? You must be joking..
I did RTFA, and understand that IBM developed the technology, but is leasing the technology to others to create product (much like NVIDIA does with its products), but my concern is that they will go the way they (and others, eg: RAMBUS) have gone before. Microchannel died because of eye popping 'we want more money for the technology' costs. If the license is reasonable, then the technology will be widely adopted. If the license is not reasonable, alternative technologies will be created to fill the void, and whether good or bad, this technology will wind up in a museum; an accidental freak side show of the information age. One thing companies are starting to learn is that if you get too greedy with new technology, it won't be viewed as a passing lane, it will be viewed as a speed bump, and a new lane will be created, perhaps of gravel instead of asphalt, but it won't be proprietary and as more people use it, it will get smoother and faster. Perhaps IBM will offer licenses at $10 million per manufacturer per year --that would be reasonable. RAMBUS wanted $100 million per manufacturer per year. Quite a bit of their technology is in the museum.
IBM no longer has anything whatsoever to do with Hard Drive manufacture, they sold that business to Hitachi years ago. IBM does sell storage equipment (quite a bit of it), but it contains drives made by somebody else.
Even if it stays more expensive, if solid state can get close (within a factor of 2, say) to the same cost for the same storage space it'll take off like wildfire given the tremendous speed and reliability advantages.
I don't think IBM is declining to actually manufacture the chips because they don't know how; rather it's because IBM doesn't have fabs large enough to make the chips in sufficient quantities. IBM's fabs are sized for CPU production, not mass-scale memory chip manufacture.
That's nice. The only nail that counts, will be the one where solid state is at least modestly cheaper for a given amount of space.
Mind you, I look forward to that nail, but until it gets here, it's not yet time to party.
For most users, disk storage is already much larger than needed. All that is needed is that an SSD have cheaper unit cost while still providing adequate storage. Even if it is still 4x the cost per byte, it won't matter because the extra storage isn't useful.
If you stick to performance available for single-digit watts, not much need for active cooling.
XML is like violence. If it doesn't solve the problem, use more.
This is pretty much how the AS/400 / iSeries works.
Am I the only one that saw the "up to 5 million write cycles...enterprise usage of 3,00 write cycles per hour" as a pretty fail example? By their own statement that's 70 days of life in that use case....
Yet another obscure company who sells licenses but doesn't make products, so they never have to pay anyone else or cross-license. Whoever heard of this "IB--" oh. Oh, them.