The Rise of Storage Class Memory (SCM)

I read this blog and would leave my first comment. I don't know what to say except NICE BLOG with a lot of information.

Jim Handy: I not only remember that article but actually had it (and subsequent events:) in mind when referencing the high-k materials. I think the 10 years was incredibly close, although I assume you would agree that we've yet to see the "introduction of a ferroelectric DRAM of a certain density." But then that article was published in October of 1999 so you still have time to be dead on the money. LoL! For what its worth (not much:)I suspect AIST and/or Prof. Ma's friends will have an array suitable for "introduction" by February's ISSCC if not December's IEDM. I don't know much about Braidwood but will look it up. I take it the goal is some ultra efficient management of the FLASH array. If it's as good as you suggest I imagine it's made for some heated debates at the Intel/MU JV. LoL! It will be interesting to see what the memory houses do in response. cb

Good coverage of Rich Freitas' talk, Steve! Interesting to see cidbarca mention Dr Ma's Yale/SRC work on FeFET for DRAM. For those who missed it, go to www.src.org/member/news/release-8-11-09.asp. Strangely enough, in a 1999 article I predicted that this would happen in ten years! eetimessupplynetwork.com/showArticle.jhtml?articleID=2905336&queryText=ferroelectric Nobody seems to doubt that there is a need for something more than what is now in production, and some technologies appear to be in the lead today, but there is plenty of time for a new contender to rise to the top. In the mean time, computer architectures are poised to change, and it looks like Intel's Braidwood will bring about the memory structure at the bottom of your figure a lot earlier than 2013, which we expect to be to the detriment of both the DRAM and SSD markets.

Good coverage of Rich Freitas' talk, Steve! Interesting to see cidbarca mention Dr Ma's Yale/SRC work on FeFET for DRAM. For those who missed it, go to www.src.org/member/news/release-8-11-09.asp. Strangely enough, in a 1999 article I predicted that this would happen in ten years! eetimessupplynetwork.com/showArticle.jhtml?articleID=2905336&queryText=ferroelectric Nobody seems to doubt that there is a need for something more than what is now in production, and some technologies appear to be in the lead today, but there is plenty of time for a new contender to rise to the top. In the mean time, computer architectures are poised to change, and it looks like Intel's Braidwood will bring about the memory structure at the bottom of your figure a lot earlier than 2013, which we expect to be to the detriment of both the DRAM and SSD markets.

Harley Burton: There was a lot of development on bubble memory back in the 1970s. Intel pushed the technology, for example. However, the bubbles live as cylindrical magnetic domains in synthetic garnet guided by little magnetic tracks laid down on top of the garnet. Electromagnetic windings around the device provided magnetic motive power for the bubbles. As I recall, a permanent magnet was also needed to provide a constant field to encourage bubble formation at the bubble generator. The strength of this field determined the size of the bubble domains. Bubble technology made little use of semiconductor process technology aside from the lithography itself. Also, the bubble devices in the 1980s were serial devices, therefore relatively slow. I've heard of no one eager to restart bubble research. Considering the device complexity, I don't expect to hear of such efforts. But you never know.

Yup, clearly a troll.

I agree with Harley Burton. Unlike what Mr. Freitas thinks, it is bubble memory that has PERHAPS the greatest chance of successfully challenging NAND Flash (or at least as good of a chance as the rest of the unproven, expensive contenders). At least bubble memory was commercialized for awhile (or so it appeared). I especially liked the warm-up period! Gave you plenty of time to think about the meaning of life and such before you started decapitating those aliens. As I said, dear Mr. Leibson, your bullshit filter is really malfunctioning. Did you read the datasheet? Did you see the write throughput? PCM will never be SCM. It is obvious for anybody who has read the datasheet and has a shred of critical, functioning brain. But go on, pump it, see what it will do to your reputation.

What about an old 1980's technology, magnet bubble memory? It seemed like it was really going for a while then RAM and EPROM got cheaper so it was discontinued. I remember it having excellent retention and access speeds significantly faster than disk. It's process was moderately expensive but it did work and I'm sure could be made economical today. I'm not capable of doing the research, but I was wondering if anyone has dusted off that bookshelf in a long while.

I couldn't agree more Steve. I start dumping salt by the bucket every time I read a semiconductor companies "views" on the topic. As to the next main stream NV commodity chip, I have only two opinions: 1) It will be determined in Asia; and 2) Within 3 years every fab producing it will be hemorrhaging red ink. But who knows perhaps God will have mercy and truly segment this business and strike down any participant who dares attempt to bridge the assigned segments. How these people can be trusted to create such important products while insisting on maintaining "The Business Plan From Hell" is beyond me. cb

There are many ways to be wrong in this world. One of the easiest is to make predictions about digital memory.

IBM's portrayal of the different technologies, if not biased, is dated. For example, RRAM's scalability and endurance have already been demonstrated at IEDM and NVMTS last year. I think the order of prefered replacement of NAND was correct at one time, but not now. Racetrack is an interesting possibility, but some form of RRAM (there are already many choices) will arrive much sooner.

Google also dredged up this excellent tutorial on SCM from SNIA: www.snia.org/education/tutorials/2009/spring/solid/PhilMills_Future_of_Solid_State_Storage.pdf

Thanks Steve Leibson. I will look at his slides but I suspect you are correct as to his categorization and use of terminology. The Japanese researchers reference their ferro memory as a "ferroelectric NAND" presumably because it uses a Flash stack structure with the ferroelectric (SBT) material as the gate. Recently, as I'm sure you know, Prof. Ma at Yale through SRC grants apparently came up with a FeFET of their own but using a DRAM process structure to form a memory cell. Their design may be cheaper in a fab but the endurance won't allow it to serve as a NV memory. But if Freitas had commented on either of these I'm sure it would have stood out. As you say ferroelectrics are not new, the AIST paper gives 1963 as the date for the initial concept of a FeFET but planar FRAM products did not go commercial until 1997. I think these ferro technologies are only gaining interest today because of their performance specs and, perhaps more importantly, because high-k dielectrics are now becoming common place on the bleeding edge of fab processing. cb

cidbarca: Freitas explicitly broke out ferroelectric RAM separately from Flash because it uses a completely different storage mechanism involving the rearrangement of atoms within molecules rather than trapped charge. Enhanced Flash for Freitas means the trapped-charge sort of memory cell, enhanced with process technology and new materials that reduce or eliminate write-wearout failure. As for Freitas making specific reference to SrBi2Ta2O9, I remember that he mentioned the old original substance, PZT (which I covered for EDN back in the 1980s) and at least one other compound by name, but I don't have it in my notes and I don't see that the Summit organizers have yet posted any of the presentation online. Freitas zipped through the eight types of memory pretty fast in his 30-minute talk and didn't get into much detail for most of the new memory types. He spent a lot of time on magnetic racetrack memory, as I recall, which is really far out and in very early R&D. Who knows if it will get anywhere. I Googled up this presentation, which contains many of the slides Freitas used: www.cs.sonoma.edu/cs_dept/events/CSC_SCM_Mar25_09.pdf

I see your blog on Handy's analysis lit quite a fire. Sorry if I fanned the flames but, excluding all the name calling, some of the comments were informational and interesting. Your effort today is as well. I was wondering whether or not R. Freitas made mention of any ferroelectric, such as SrBi2Ta2O9, as one of the "new materials" under the "Improved FLASH" category? Or whether or not ferroelectric polymers were mentioned among the resistive "organic materials" in the RRAM category? There has been quite a bit of "disruptive" work on ferroelectrics for memory and logic applications at AIST over the last 8 years that may be of interest: www.aist.go.jp/aist_e/latest_research/2008/20080624/20080624.html www.jstage.jst.go.jp/article/elex/6/12/831/_pdf cb