Intel touts NAND-killer breakthrough
Big changes for phase changes
Intel and chip-tech house Numonyx unveiled a new technology on Wednesday that the companies claim will enable non-volatile memory to break through NAND's 20nm barrier and scale down to process sizes as tiny as 5nm - and do so cost-effectively.
What's more, the resulting stacked memory arrays could potentially usurp current DRAM and NAND-storage chores. As explained to reporters this morning by Al Fazio, Intel Fellow and director of memory technology development, this could even allow system designers to "collapse some of the DRAM and some of the storage attributes of solid state into one memory class."
In simple terms, that would allow the bringing together of DRAM and storage into one high-speed, high-bandwidth architecture. But such a leap is a long way off - products based on today's announcement won't appear for "many years," according to Fazio.
The breakthrough described by Fazio and Greg Atwood, senior technology fellow at Numonyx, is an advancement of phase-change memory (PCM) technology called PCMS (phase change memory and switch) that creates thin-film memory cells and their controlling thin-film selectors out of the same basic chalcogenide-family material and marries them together in a crosspoint architecture.
This new phase-change technology may some day blend your memory and storage into one happy family
This new thin-film selector, dubbed an Ovonic Threshold Switch (OTS), allows the stacking of multiple iterations of the memory/selector layers on a CMOS base to create high-density, high-bandwidth PCM memory.
Well, such stacking is the goal. Wednesday's announcement was of a working 64Mb, single-layer version of the new memory architecture that will make its official debut in a paper to be presented at this December's 2009 International Electron Devices Meeting in Baltimore, Maryland.
But multiple layers of the new devices are on the drawing board. As Atwood said, "The first layer is the hardest layer."
The advantages of PCM over current NAND memory are legion, according to Fazio and Atwood. For one, the electron count in a NAND cell shrinks each time the process size shinks - and it's currently down to "a handful," according to Fazio. What's more, internal voltages in a NAND cell run to over 20 volts, creating pesky interference from the resulting electric fields.
PCM, on the other hand, is read not by a charge but by the resistance of its state - either crystalline or amorphous. What's more, PCM bits are "RAM-like" in that they can be changed individually, and not merely in blocks as is required by NAND, according to Atwood.
The advantages of RAM-like capabilities are clear - at least to Fazio: "You have a memory technology that looks like memory - in other words, hardware can do a load/store because it can act on a small chunk of data with a low latency, yet it's non-volatile so that it has the non-volatile aspects of storage."
Sounds like a win-win to us - especially as Moore's Law pushes those process numbers down into the 5nm range, and the technology announced today is developed into cost-effective, high-density packages.
Fazio and Atwood have no doubt that it will. ®
Look, you whiners out there that want the vendors to "just show me the money, sonny" obviously have never worked at the coal face. I have, for virtually my entire career (22 years and counting). It takes time and timing to shepherd a new technology from the lab all the way to the retail store. And big companies and governments share the common trait that they move like snails when it comes to change, so its a good thing the bleeding edge takes time to advance -- it gives folks like me time to prepare the culture and infra to accept the new technology. I rely on El Reg (among many others) to bring me the news of where all these advances are -- good job, Reg!
So quit yer yapping, kick back, soak in the future-looking buzz, and imagine what YOU would do with PCM memory. THEN, TELL INTEL!! How do you think they know which technologies to get behind, which to put on a back burner to cook a bit longer, and which to shelve? Yeah, partly the technology itself tells them...but in great part, its *US*, their potential customers, that tell them. [And for the record, I've been telling them we need PCM memory for a few years now.] Its up to us to look over the technology, understand how it works, see if we can find any potential flaws or drawbacks, imagine how we'd use it, and tell the vendors who are trying to bring these technologies to market.
If you don't -- if you just sit back and kvetch about the news -- then you really can't complain about the speed or capacity of your technology, now can you?
heard this one before
Numonyx previously used BJTs to do PCM. They introduce OTS here for multilayer but actually only demonstrating single layer. Having studied OTS I can say they are going out on a limb. Just to convince industry to use PCM instead of flash is difficult . Now we also have to rely on OTS instead of transistor. Two immature technologies... bad move.
> Now all we need are OSes which would work with this architecture.
something along the lines of single-level store? as in Multics or OS/400?
Indiana jones and the search for the holy grail of memory...
PCM is good but it isn't the holy grail of memory, but still its better than Flash in a number of ways, (although Flash wins in a few other ways) but its not going to replace RAM ever. PCM still wears out. In theory PCM can take a lot more writes than Flash but its not like RAM which allows unlimited writes. Still PCM does give far higher numbers of writes combined with the fact PCM can also address individual bits which greatly helps to increase its storage lifespan. Flash works ok but its got a lot of limitations. One advantage Flash has is that it can allow multi-bit (Multi-level) storage which increases its density (but partly at the expense of reducing the number of reliable writes).
But I don't see how Flash is going to be displaced by any other memory technology until it offers better storage densities than Flash. (Until that point any other memory technology will always remain very niche market memory and so some people will unfairly laugh at it for not being mass market). The point Flash memory looses the race to better storage densities won't happen until we are approaching much smaller scales as its hard to scale Flash down. (Intel aims to start 11nm by about 2015-2016). So we are looking at a few years at least where Flash is unchallenged in storage.
The only 2 technologies I've seen so far that could in theory over take Flash in memory density (in the next few years) are Memristors and maybe just maybe PCM, but both still wear out so none are RAM style replacements.
I am however unclear on the potential of FRAM. I'm not sure if they wear out eventually. FRAM has however got the potential to be very fast, in theory it could beat DRAM so if it doesn't wear out thats one to very much watch?
As for MRAM this one is even more unclear to me. It in theory doesn't wear out! and its as fast as RAM and it has similar densities to DRAM but unfortunately both DRAM and MRAM don't have the high density of Flash. Problem is that means MRAM doesn't yet have a market. It can't replace DRAM as it would require a redesign of computers which won't happen for the vast majority of computers (so MRAM stays a niche market) and MRAM can't replace Flash as it can't match its storage density (so again MRAM stays a niche market). MRAM seems to be currently trapped by market forces at the moment. I think one big hope for MRAM is how it can also save power. That could open up its potential into many markets as both a DRAM and smaller Flash sizes replacement in one chip. Like maybe the phone market. By replacing both DRAM and Flash and saving power while on and off its got a lot of potential. It could also do very well in embedded applications for the same reasons as it can act as both DRAM and some storage space.
If they could crack the density issue with MRAM then it would become the holy grail of memory. That would suddenly make it extremely big business.
From a marketing point of view, in the future I can't help thinking having instant on PCs would be a major selling point to most people who hate waiting for things to boot up. Its not hard to imagine a future where the idea of waiting for a computer to boot up would becoming as old fashioned as listening to a modem is to us now. (Imagine 2 people with computers both switching their machine on at the same time and one instantly goes ping and is ready to use, while the other computer they have to sit and watch for a few minutes slowly getting ready to use). I suspect most people would end up laughing at boot times as very old fashioned. But I think its going to take a big company to make their own high density MRAM for their own Sub-notebook to really shake up the market to really kick start MRAM sales, but if they do I think they could end up making a fortune from MRAM. My guess would be companies like Samsung and Sony could do it. (My money is on Samsung. :)
It is what I am waiting for
The unified memory - get rid of all this copying data back and forth, don't bother 'loading' an app into RAM, just run it where it is, modify a few bytes her and few bytes there, only duplicate when you need to.
Now all we need are OSes which would work with this architecture.