You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
I don't know if you have seen the latest news, Intel and Micron have announce a non-volatile memory 1000 times faster than NAND memory. Peering closer at 3D XPoint memory: What are Intel, Micron up to?
Production is to start next year. Is it possible that this would make NAND memory obsolete?
This is a new technology, so it remains to be seen how much it will shake up the memory area. If Intel/Micron can demonstrate high volume production that is cost competitive / density competitive with NAND flash memory, then it does stand a chance of succeeding. Another factor here we need to consider is technology-related. There are competing approaches for ReRAM or RRAM (aSi, Pt/Ti:NiO/TiOx/Pt, etc.). It is possible that one of these other technologies ultimately emerges as the better technology. Still another is business-related. Crossbar is working with some of the foundries to create this type of memory as well. Perhaps their positioning will work better in the long run. Finally, this type of memory is not as fast as DRAM, so DRAM will still be necessary for certain areas.
Crossbar is slated to be delivering IP for their RRAM for embedding in logic circuits by the end of this year. There will be a few flavors of RRAM coming out soon that have been in development for a while. These processes should be the beginning of 2D technology. Crossbar has been working with TSM for about two years having them develop the production process.
Intel was up on the stage and they would not have put their name on this if they did not think it was disruptive. After all they know what is possible with a new form of memory.
Here is a link to an interesting article that lays out some interesting ideas about what could be in store:
Oh and regards Crossbar and the other IP only NG memory companies I think we are a long way from seeing any real devices. New forms of devices need tons of real live fab work to prove out. The IP/foundry model won’t work. It’s not like you can validate the designs on standard design/simulation packages. They don’t exist!
They said that this was NOTE ReRAM, but something different, and that it was between Flash and DRAM speeds. I'd say it could replace SSD which is a huge market to displace.
It also looks like Sandisk has similar technology in the pipeline. Due to cost structure, the adoption of this technology is expected to take a while. Having followed this technology for a while, we can expect to see new developments and flavors come out soon.
Agree, many new flavors. SSD in the data center seems a very good application but then so does on processor storage. assuming this is BEOL compatible which I think it is. If so Intel can disrupt the mobile processor space along with several others.
Being memory there is no doubt that the other big players will have a product but I think Intel/Micron will have some time to earn first mover returns.
I highly recommend the seeking alpha purple swan article I linked in my first post on this subject. Mr Breezy has some very good info. I'd go long on Intel.
ippisl, thanks for linking the Mr Mick's article. It was very informative. About three quarters of the way in it references Stephen Breezy's Seeking Alpha article. For those interested in IP re this new device I highly recommend this article.
ippisl Thanks for the link! It also mentions a little bit about the optical backplane that is related to what I do. I've been looking closely how they'll use their demoed 100G Si photonics link. The implementation has been delayed several times (just like some of their process nodes ). It seems to be in the form of optical PCIe in their closed OmniPath ecosystem that Intel tries to increase its value by push it into datacenters, their only foreseeable fast growing BU.
ippisl Thanks for the link! It also mentions a little bit about the optical backplane that is related to what I do. I've been looking closely how they'll use their demoed 100G Si photonics link. The implementation has been delayed several times (just like some of their process nodes ). It seems to be in the form of optical PCIe in their closed OmniPath ecosystem that Intel tries to increase its value by push it into datacenters, their only foreseeable fast growing BU.
Fred, yes and probably just in enterprise data centers which are smaller and prefer a modular system. Intel chose a full proprietary scheme with even customized optical connector (from USConec) and fiber (from Corning)! It's hard to penetrate football-field-size mega data centers like many ones built by Google, Amazon, Microsoft, Facebook, Apple, etc. These players always try to use off-the-shelf silicon as well as optics (although many are sub-spec for cost reduction) and built the data center racks either in-house (e.g. Google) or using custom white-box design (e.g. Facebook-lead Open Compute Project).
The proposal of using an optical backplane has been around for years but only recently becomes close to adoption when the year the 25Gb/s baud rate (per lane) finally arrives both on-board and in interconnect fabrics in data centers. Although I call it "backplane" it goes beyond just backplane form factor and is likely first implemented in the form of so-called embedded optics module (EOM) between main board and/or mezzanine cards. IBM once implemented Avago's 120Gb/s (12x10Gs) Micropod EOMs in their BlueWater (P775) HPC systems and consumed several hundred thousands units (which is not a small number for high end optical modules). But there has been nothing quite exciting since a couple of IBM's bulk orders several years ago. It's only until recently that EOM emerged again because running 25Gb/s (and dense channels) with copper really causes issues like reach (tens of centimeters), power consumption (from re-timing), and EMI.
For Intel, will all those technologies aimed at the server(XPOINT ,altera, fiber backplane, eAsic customization for microprocessors, 3d assembly ) - enable great growth for intel ? or are they mostly to protect against ARM , now that fab leadership has weakend ?
At least the Xpoint technology is just a new type of memory structure, not really aimed at ARM or any company in particular.
For Intel to compete with ARM they would have to design a processor core that is power efficient, have a popular software stack and be something that they would license, just like ARM to propagate across multiple industry segments. Intel however has a history of not wanting to license their core IP, so I don't see Intel successfully competing with ARM any time soon if they maintain their present strategy.
The last time that I recall Intel licensing their core processor IP was to IBM, but that was after IBM had purchased some 13-14% of INTC stock as Intel was in grave financial trouble in the 1980's.
Ippisi, I think the datacenter/HPC market is what now Intel is focusing on and Intel wants to sell more than just server CPUs. All these things you mentioned more or less have something to do with this. For example, part of the reason of the Altera acquisition may be because mega datacenter operators such as Microsoft began to use FPGA on their servers along with NICs to reduce CPU's workload so as to make CPU to concentrate on compute instead of communication. It might represent a trend and Intel want to be part of it.
I think, currently with >90% market share, there is no real threat for Intel in this market segment. IBM's PowerPC may always be there but only as a niche, and it seems to take a long time for 64bit ARM server chip to gain any tractions. So the main thing for Intel now is to gain as much value as possible by integrating as many things (optics, FPGA, Xpoint, etc. ) into their OmniPath or some other proprietary ecosystem.
). It seems to be in the form of optical PCIe in their closed OmniPath ecosystem that Intel tries to increase its value by push it into datacenters, their only foreseeable fast growing BU.