News & Analysis
Comment
Helicopter
Micron, on their website already has 2x nm SLC NAND capable of 300k cycles and ...
greenpattern
It doesn't bode well for PCM if even Mr. Doller considers it "expensive," but if ...
Micron confirms support for phase-change memory
Peter Clarke
7/30/2010 1:14 PM EDT
LONDON — Micron Technology Inc. (Boise Idaho) has confirmed its support for the phase-change memory product line and R&D inherited when it acquired Numonyx BV in May 2010. But a company executive declined to give further details of its technology roadmap or how it plans to scale the technology below the 45-nm node.
An article by Micron senior fellow Greg Atwood contributed to EE Times, entitled The evolution of phase-change memory, has excited considerable response as well as a counterpoint article entitled Phase-change memory: A rebuttal of Micron's article.
When asked if Micron continued to support phase-change memory Ed Doller, vice president and chief memory systems architect with Micron and formerly chief technology officer with Numonyx, told EE Times: "Absolutely I can confirm that." He added that 80 to 90 percent of the staff that report to him are working on phase-change memory.
Phase-change memory has shown promise as a potential non-volatile replacement for flash memory and DRAM but the technology has been slow to make it out of the laboratory. As the situation stands PCM is behind the rival technologies in terms of both memory capacity in bits and the cost of manufacture. This has given rise to speculation that the technology may be a curiosity that ultimately fails to impact the market. Micron has one phase-change memory product line that customers can buy; the Omneo 128-Mbit device implemented using a 90-nm process technology but Numonyx, and now Micron, have yet to produce a promised follow-on 1-Gbit memory using a 45-nm process technology.
When asked if Micron would respond directly to the technical questions posed within the Internet pages of EE Times, Doller said that Micron has put some technical information in the public domain through the presentation of papers at conferences such as the International Electron Devices Meeting and that other technical material it shares with companies under non-disclosure agreements.
Doller did not disagree with the statement that other memory technologies are already down at 3x-nm and heading for 2x-nm so that for PCM to have a future it must compete at those geometries and beyond.
Doller responded: "There will be challenges but we are on a path to scale the technology further than any of the incumbents, by which I mean NOR, NAND and DRAM." He declined to discuss detailed solutions to the challenges to the scalability of PCM. These could include the non-scaling of current density, electro-migration and thermal cross-talk, amongst others.
When asked about the 1-Gbit device, Doller said: "We will be sampling a 45-nm 1-Gbit to customers this year. Volume production is in 2011. There is no doubt in my mind about this."
Doller said that PCM is targeting mobile phone applications as well computing applications. A different emphasis apparent in Atwood's article was due to the fact that the article was only addressing computing platform applications where Doller said the company does expect design wins in storage applications to come first.
"Is there value in putting expensive PCM in there? When you think about latency, yes," said Doller outlining that magnetic media have delays of the order of milliseconds compared to microseconds for NAND flash solid-state memory, which compares to nanoseconds for PCM. "There is also a case for endurance. NAND endurance is reducing rapidly as we scale," asserted Doller. There continues to be an opportunity in mobile handset memory, said Doller.
When asked about what geometry node Micron would target for the next device after 45-nm and what progress was being made developing a next-generation phase-change memory IC, Doller said: "We are working on what comes next," but declined to comment further.
Related links and articles:
www.micron.com
Numonyx 1-Gbit phase-change memory delayed
An article by Micron senior fellow Greg Atwood contributed to EE Times, entitled The evolution of phase-change memory, has excited considerable response as well as a counterpoint article entitled Phase-change memory: A rebuttal of Micron's article.
When asked if Micron continued to support phase-change memory Ed Doller, vice president and chief memory systems architect with Micron and formerly chief technology officer with Numonyx, told EE Times: "Absolutely I can confirm that." He added that 80 to 90 percent of the staff that report to him are working on phase-change memory.
Phase-change memory has shown promise as a potential non-volatile replacement for flash memory and DRAM but the technology has been slow to make it out of the laboratory. As the situation stands PCM is behind the rival technologies in terms of both memory capacity in bits and the cost of manufacture. This has given rise to speculation that the technology may be a curiosity that ultimately fails to impact the market. Micron has one phase-change memory product line that customers can buy; the Omneo 128-Mbit device implemented using a 90-nm process technology but Numonyx, and now Micron, have yet to produce a promised follow-on 1-Gbit memory using a 45-nm process technology.
When asked if Micron would respond directly to the technical questions posed within the Internet pages of EE Times, Doller said that Micron has put some technical information in the public domain through the presentation of papers at conferences such as the International Electron Devices Meeting and that other technical material it shares with companies under non-disclosure agreements.
Doller did not disagree with the statement that other memory technologies are already down at 3x-nm and heading for 2x-nm so that for PCM to have a future it must compete at those geometries and beyond.
Doller responded: "There will be challenges but we are on a path to scale the technology further than any of the incumbents, by which I mean NOR, NAND and DRAM." He declined to discuss detailed solutions to the challenges to the scalability of PCM. These could include the non-scaling of current density, electro-migration and thermal cross-talk, amongst others.
When asked about the 1-Gbit device, Doller said: "We will be sampling a 45-nm 1-Gbit to customers this year. Volume production is in 2011. There is no doubt in my mind about this."
Doller said that PCM is targeting mobile phone applications as well computing applications. A different emphasis apparent in Atwood's article was due to the fact that the article was only addressing computing platform applications where Doller said the company does expect design wins in storage applications to come first.
"Is there value in putting expensive PCM in there? When you think about latency, yes," said Doller outlining that magnetic media have delays of the order of milliseconds compared to microseconds for NAND flash solid-state memory, which compares to nanoseconds for PCM. "There is also a case for endurance. NAND endurance is reducing rapidly as we scale," asserted Doller. There continues to be an opportunity in mobile handset memory, said Doller.
When asked about what geometry node Micron would target for the next device after 45-nm and what progress was being made developing a next-generation phase-change memory IC, Doller said: "We are working on what comes next," but declined to comment further.
Related links and articles:
www.micron.com
Numonyx 1-Gbit phase-change memory delayed
Numonyx set to discuss potential PCM showstopper
Micron-Numonyx deal: What analysts are saying
|
|
|
|
|
Navigate to related information
Volatile Memory
7/30/2010 3:08 PM EDT
It is amazing that Mr. Doller is still being taken seriously. As the author of this article had noted previously, Mr. Doller promised that mass production of the 128Mbit 90-nm PCM chip will start in 2007 ( http://www.eetimes.com/electronics-products/other/4091980/Intel-to-sample-phase-change-memory-in-1H-2007 ). At that time, he also claimed that the "128-Mbit phase-change memory had demonstrated 100 million read-write cycles of endurance" and that the "phase-change memory gets pretty close to Nirvana," and that the chip would "start to displace some of the RAM in the system."
Fast forward to today. No product on the market uses any PCM chip. The 128Mbit 90-nm chip is specified to just "over 1 million write cycles" (see page 1 of the datasheet), and the thing is slow as molasses (even in page mode it writes at a typical 0.5MB/s, and could be as slow as 0.2MB/s, according to page 71 of the datasheet) - it can't even compete with NOR & NAND, much less with SRAM or DRAM.
So much for the Nirvana...
Micron, having signed an exclusive, royalty-bearing license with ECD for the patents in the area of phase-change memory in 1994 (years before Mr. Doller got involved), after 10 years of trying, declared in 2004 that they didn't see how the technology could "scale effectively with Moore's Law and justify commercialization." ( http://www.eetimes.com/electronics-news/4123177/Micron-drops-MRAM-phase-change-mulls-other-memories ). Micron's statement was correct in 2004 and is still correct today. Why would Micron allow their reputation to be ruined now is quite puzzling.
Sign in to Reply
selinz
7/30/2010 6:24 PM EDT
Well, curiousity got the best of me on this one... I substituted "TV time" looking up patents and papers in the area. As far as I can tell, it appears as though there is "theoretical" support from IBM, Samsung, Micron, Intel, Infineon, ST, BAE, and Hitachi that this is the most likely candidate for "storage class memory." This is pennies/GB stuff that will combine 3D stacking, multilevel cells, and "litres" of storage space. If you look at the trend of "user devices," it is in accessing information rather than creating new information. So having efficient read capability is more important that write speed. IBM almaden has a nice presentation that culminates talking about linearizing the human brain's 20 billion neurons into approximately 2.5 petabytes of states. In other words, approximating the brain is more likely to happen with a data-centric paradigm rather than a compute-centric. The 2007 speculation was that they will have a human scale brain simulation by 2017.
In my household, our family of 6 is constantly generating data (pictures, music, videos) that is stored once and read many times. PCM is not destined to take the place of compute-centric application (i.e. dram on motherboards) but rather long term storage and fast and random read. We went from generating about 1Gbyte of data from 1983 to 2000 to about a Tbyte/year today. Most of this is stored and never rewritten.
I think that if we step back and think about (or guess is probably more accurate) how future storage will be used, how much we'll need, and how much it costs (in space, money, and access time)then perhaps the specs that are cited in the above comment may not be the only important things to focus on...
Sign in to Reply
Volatile Memory
7/30/2010 7:46 PM EDT
selinz: I suggest you substitute a bit more "TV time" and read about bubble memory ( http://en.wikipedia.org/wiki/Bubble_memory ). "By the mid-1970s practically every large electronics company had teams working on bubble memory. By the late 1970s several products were on the market, and Intel released their own 1-megabit version, the 7110. In the early 1980s, however, bubble memory became a dead end with the introduction of higher-density, faster, and cheaper hard disk systems. Almost all work on it stopped."
So, yes, despite all the professed interest by practically every large electronics company, phase-change memory faces exectly the same situation as bubble memory - Flash (which was invented circa 1980, or about 10 years after phase-change memory) currently has 500x better density, 20x better speed, and is 150x cheaper (per GB capacity). That is why no product on the market uses any phase-change memory and that is why phase-change memory will never be commercialized in volume.
If your household constantly generates data, I wonder how you would feel if you have to use a new chip for every two new pictures and that "pleasure" will cost you $2 per picture (because a 128Mbit phase-change memory chip stores just 16Megabytes, which can barely fit two pictures from a 10 megapixel camera in raw format). Oh, and yes, it will take 10+ seconds to store each picture. But, hey, according to Mr. Doller, the latency will be superior!
Some future of storage!
Sign in to Reply
resistion
7/30/2010 9:17 PM EDT
Some companies working on MRAM. Would that be a better technology to use than PCM?
Sign in to Reply
greenpattern
7/31/2010 7:08 AM EDT
Cell phones emit magnetic fields. I doubt MRAMs are appropriate in that environment.
Sign in to Reply
ece-professor
7/30/2010 9:45 PM EDT
MRAM has no way to compete with NAND flash in terms of storage density or GB/$, though it may find its role as embedded memory to replace eFlash or eDRAM.
A recent very good PCM survey paper from IBM says "if PCM fails, it will be on a cost basis ...", with which I totally agree. Meanwhile, with the help from strong coding and signal processing and other system-level tricks, it should not be a big surprise if NAND flash will be scaled to ~10nm, and we could imagine how difficult it would be for PCM to compete with NAND flash on a cost basis. Lastly, as storage capacity increases, the touted endurance advantage of PCM will become very much irrelevant in most applications.
Sign in to Reply
resistion
7/30/2010 11:01 PM EDT
The trouble is system-level fixes add system-level costs which are not benefited by scaling (they add area outside the array). Maybe MTJs don't need so much. Can PCM process cost be lower than NAND?
Sign in to Reply
selinz
7/31/2010 1:40 AM EDT
History is often a good indicator of the future performance, but not always--particularly when the history is 40 years old. No TV time needed for bubble memory as I was an intern with IBM on Cottle Road in the early 80's when they made the decision to shut down the bubble program.
As far as PCM failing on cost basis, that is pretty much why anything fails. That's why bubble memory failed. A garnet wafer with the Liquid phase epitaxy film was just too expensive... (even though we had demonstrated 0.4 micron resolution using phaseshift masks) Hard drives have fought off nearly every technology in terms of cost for 30 years. A 500gb drive for $50 at fry's in todays paper suggests that it is still doing it. That's good history...
Sign in to Reply
selinz
7/31/2010 1:41 AM EDT
500gb notebook drive, that is...
Sign in to Reply
greenpattern
7/31/2010 7:17 AM EDT
The article rightly indicates current density, electromigration, and thermal crosstalk as important issues often ignored. The atoms are supposed to be heated to the point of displacement, right?
Sign in to Reply
Robotics Developer
8/2/2010 10:51 AM EDT
Having worked in the mass storage division of a computer company I can say: "It is all about the cost per MByte". We had higher performing drive systems we owned/developed, but in the end, the lowest cost off the shelf drives displaced the much higher performing but very expensive drives. Cost rules in most of the consumer applications (cellphones, TVs, games, etc.). If PCM can be cost effective it will be successful, if not, then it will be an interesting curiosity.
Sign in to Reply
greenpattern
8/2/2010 12:10 PM EDT
It doesn't bode well for PCM if even Mr. Doller considers it "expensive," but if we carry the cost per bit argument further, flash won't win out over HDD. But there are other considerations indirectly related to cost, like power consumption.
Sign in to Reply
Helicopter
8/2/2010 11:41 PM EDT
Micron, on their website already has 2x nm SLC NAND capable of 300k cycles and the future story continues to look good for NAND.
It looks like due to its large die size and temperature limitations, PCM is destined for niche markets
Sign in to Reply