Seems an optimistic timeline to 'reinvent' memory. They have perhaps forgotten about 'bubble memory' which was going to 'eliminate disks' in 1982.
From the article : "As reported by The Register, at a recent conference in Oxnard, California, HP’s Stan Williams said that commercial memristor hardware will be available by the end of 2014 at the earliest."
So basically we'll get to see real devices perhaps at the end of 2014 (I'm guessing closer to 2017 but we'll see) And we need a couple of years of building/using/repairing them before we see wide spread design wins, then another year before 'memristor' enabled devices hit the market and are or are not competitive.
That said, I'm rooting for them to be successful, flash is in a bad way at the moment with feature size being a hard limit on cell lifetimes.
Bubble memory was doing just fine until a Nobel-winning physics breakthrough allowed the existence of high-density hard drives. If memristors do as well as bubble memory, there will have to be a serious fluke for it not to replace everything on the market. https://en.wikipedia.org/wiki/Bubble_memory#Commercializatio...
Ah thanks for the memories (I was at Intel when they rolled up the bubble memory team). But if the Wikipedia page can be trusted, its a good example of the timeline of a disruptive technology. So 10 years from first devices to a chance to really disrupt everything. That would make 2018 the year to watch for memristor memory.
Swings and roundabouts - Intel did much the same to promising processors, with their x86 juggernaut. With unlimited cash reserves, you can put lipstick on a pig.
Hey, be nice to MRAM, it's useful stuff. It has horrible densities, but its as fast and reliable as you could please. I've certainly used it in applications where there wasn't anything else that could foot the bill.
Some of the controversy is about priority, which may not matter so much; I care less about whom I get massive on-chip non-volatile storage from than that I get it at all. But that too is under dispute (e.g. "Myth #3" in the second link above). So it's a little distressing to see signs of vaporware from HP at this point. I really want this!
Also, the OP links to a post (http://www.theregister.co.uk/2012/07/09/hp_memristor_and_pho...) in which Williams says something odd: "We're not going to make money off these chips. We are going to make money by building cool systems utilising these chips." If his memristor claims are true, the chips themselves could hardly be more of a game changer – worth billions to put it mildly.
> Historically, electrical circuits were crafted with three basic building blocks: the capacitor, the resistor, and the inductor. But in 1971, University of California at Berkeley professor Leon Chua predicted the existence of a fourth: the memristor, short for memory resistor.
> Then, in May of 2008, HP announced that it had actually built a memristor, thanks to HP Labs Fellow R. Stanley Williams and others working in the company’s research arm.
Here's a great presentation from R. Stanley Williams[1], via prior discussion at HN[2].
What I take from this presentation is much, much more than lots of memory. Apparently, these things can also do processing, like an FPGA. The result would be a whole system on a chip, only the design is almost entirely pushed up at the "software" level.
To me, that's even more game changing than a mere petabyte on a square inch. "Specialized" hardware is made easy. Complete re-purposing becomes possible. Most hardware compatibility issues just go away. And of course, it will be much, much easier to experiment with novel and totally crazy architectures.
Oh, and they say it can re-configure itself on the fly, very rapidly. So, it's not just easy innovations on hardware. It's metamorph hardware. Imagine for instance code that compiles down to logic gates, instead of very high level assembly code, possibly on the fly. Or, re-allocating hardware resources by the second, depending on your needs.
It's an exhilarating vision, because it would get a whole lot of accidental limitations out of our way and allow us to build the essence of a design. A lot of complexity is due to the impedance mismatch between what goes on at the lowest level vs. the highest. To be able to design the machine itself as part of the system would be a huge breakthrough – the sort of thing that, if it happens, people may look back and wonder how anybody did anything before.
I get excited thinking about it. Unfortunately, we still don't have any evidence that it's real. Read carefully, the article rather suggests the opposite.
> “Our partner, Hynix, is a major producer of flash memory, and memristors will cannibalise its existing business by replacing some flash memory with a different technology,” he said. “So the way we time the introduction of memristors turns out to be important. There’s a lot more money being spent on understanding and modeling the market than on any of the research.”
Surely if they wait too long purely for profitable reasons the market will punish them by someone else beating them to it?
As bad as that sounds, I don't think that HP is currently making much money off flash memory, so they still have an incentive to get the memristors to market soon. They just have chosen to make enough concessions to secure the cooperation of a DRAM manufacturer so that HP doesn't have to build the whole memristor business themselves. (The way HP's been managed in recent years, I wonder if they would be able to get and hold on to enough capital to mass produce memristors and bring them to market without help.)
But HP would never attempt to maximize for profit right? The memristor is going to be its gift to humanity in the same way they've kept ink-jet and toner prices right above their costs.
I think that statement about spending more (time or money?) studying the market than doing research is pretty sad.
That passage leapt out at me too. In my simplistic geek mind this is totally arse-about-face, like the tail wagging the dog. Are these companies more interested in money than what they're actually doing?
> Are these companies more interested in money than what they're actually doing?
Their [HP] interest in what they're actually doing is demonstrated in their investment in research that has enabled memristors. Commercialization, what the quote was about, is obviously more about money. Money that eventually ends up funding more research like this.
Of course you're right, but surely there's a balance to be found between delaying the launch of this product vs. maximizing profits. If the company involved is more interested in money they'll tend to err towards delaying release, if they're more interested in the impact of their product they'll tend to release sooner. Clearly, at a societal level, we'd prefer to have this product sooner, so we have to constantly question the leverage that patents afford large corporations and question their methods and motives.
Aficionados of weird co-incidence might be interested to learn that the researcher who discovered the memristor, Leon Chua, also has a daughter who is rather famous for having written "Battle Hymn of the Tiger Mother" (http://en.wikipedia.org/wiki/Amy_Chua). Leon Chua is also an enthusiast of Stephen Wolfram's NKS, and spoke at one of the early NKS conferences.
They should starting to bring out developer packages asap, because memory is not the only thing you can do with it as you can see from the presentations and papers. I find the stuff it can do BESIDES memory actually more interesting and really would like to get experimenting with it.
Even if the base circuit elements are made of memristors instead of transistors, the traces on top of them still need to be laid out in the traditional manner. So the fixed cost of bringing even a small memristor device to market is millions.
Memristors are not some magic stuff that suddenly makes FPGAs as efficient as ASICs, nor do they make then dramatically cheaper.
I'm not talking about magic; that's why I would like to see some (even expensive) developer version (as this is how new tech these days usually starts right?) come from HP. HP has the millions I would imagine and they know/see the potential (even though a lot of investment needs to go in to harvest that potential).
I really have no idea what you even mean with a "developer version". Memristors are just circuit elements. They are laid out in the billions on a wafer and connected together with copper interconnects -- just like normal transistors or the floating gate ones used by flash. They could give devs early memory devices or something that uses memristors for logic, but other than having the word memristor on the top, they would not differ from similar devices built out of transistors and silicon.
Well, the interview with the scientist from HP clearly states software would have to written very differently to make full use of this technology. So I want to see how that works. You seem to know a lot more about it, so what am I missing?
Instead of real hardware, I wonder if a mock up virtual device acting as it's making use of memristors could work for experiments on how this could affect software.
So then memristors remain "in the future". sigh. I was naiive enough to believe them a year ago when they said "it'll be on the market in 18 months". I was really looking forward to that being true and ditching RAM in six months.
It may be 18 months in the RAM world but in memresistor world it is as long as you like and still holding the 18 months value :).
That all said you are right and sadly come 2014 it will still not be available for your PC to replace your RAM as it will take that long again to define a standard and then that time again to have chipsets that support said standard.
So for a consumer PC, i'd say 2017 is when you can look at replacing your RAM and by that time PC's will probably be reduced to devices were you can't change the RAM and as such you wont know what you have got inside. But who knows for sure and if they do then they would proabably get done for insider share dealing just by telling people.
Only think we know for sure is the here and now and that any dat that has no product release date set in stone is a date that has not had the engineering factor of x2 and is a marketing factor of x.5. IE Engineers double the amount of time it will realy take and markting will half the amount of time it will realy take. Even then Engineers thesedays need to realy use x3.
Still - once we see a prototype working in a form factor we can identify with and running windows or a comercial OS of any form, then and only then can we feel that it will be available within 2 years. Anything else is pure marketing by people who will be working elsewere in years time.
Though I would love to be proved wrong on this, truely.
You're right, I had my techs backwards. I thought they were targeting RAM first, then flash second. It's actually the other way around. Flash was supposed to be dead next year, DRAM in 2014, and SRAM in 2015 (or somewhere around there).
Gotcha. I didn't even know they were targeting RAM at all. Can't wait until storage gets cheaper and faster. IO is the worst part of the cloud and this improvement could make working with a cloud computer much better.
I'm more surprised at how surprised so many people seem to be at this. This isn't a college student building a music-sharing app here. This is a huge corporation preparing to deliver disruptive technology. Naturally they're doing enormous amounts of market research.
But it doesn't make any sense. Why not spend those enormous amounts in actually producing and shipping it before anyone else does? Market research is sunk cost and doesn't prevent failure.
Why not spend those enormous amounts in actually producing and shipping it before anyone else does?
Because you might be able to make more money with a slow introduction rather than a fast one. Restricting supply means you don't cannibalize the existing flash memory market, and can probably still make huge profits from people who must have the best possible performance.
Also, over-investment in production facilities is a huge risk.
Naturally they're doing enormous amounts of market research.
Does this pass a sanity test? What they're talking about would be the biggest breakthrough in computing in a generation. The question is not whether there's a market for it. The question is whether it works.
Memristors have "been around" in laboratory settings for a few years now. The research is done, now it's time to figure out how to market it. Seems pretty reasonable to me. This is actually very exciting to me, because the concept of a memristor, that of a large amount of small devices storing chunks of a larger piece of memory, is supposedly the same or a similar way that your brain's memory works.
The research is very far from done.
Disclaimer: I'm not a physicist but at the lab the guy to my left is one and he's doing a PhD about memristors. I have an experimental device in front of me with wires coming out and stuff, it's based on some titanium alloy.
They basically don't know why it works. They have some theories, like conductive ion micro channels forming in the material or something like that. They have two mathematical models of simulation, completely different of each other, and each one "works" under different set of conditions, but unlike transistors, flash memory and other components, the underlying effect that make most memristors work is currently unknown or theoretical. I was quite suprised because they don't tell you this in the news.
I'm sure this small detail will not stop HP or Intel from selling it anyway.
Edit: "the concept of a memristor, that of a large amount of small devices storing chunks of a larger piece of memory" <-- this is not the concept of a memristor. Not even close.
I think memristors are a real contender for future devices; the darpa synapse reseaarch has shown that it's already a big backbone for future ai devices.
I wonder if the first uses of RRAM, when its still fairly expensive and low density, will be as a coalescing cache in front of big blocks of flash memory? Or maybe just for the page table.
“Development costs at least 10 times as much as research, and commercialization costs 10 times as much as development. So in the end, research — which we think is the most important part — is only 1 percent of the effort.”
This just means that it's easy to get something working in a lab but it's hard to make a product that you can mass produce. In non-voitile memory a flash replacement is always 2 years out, just like cold fusion is always 10 years out.
That was my take, too. I've been hearing about commercial memristors for at least a couple decades now, and they're forever on the cusp of showing up on my desktop.
From the article : "As reported by The Register, at a recent conference in Oxnard, California, HP’s Stan Williams said that commercial memristor hardware will be available by the end of 2014 at the earliest."
So basically we'll get to see real devices perhaps at the end of 2014 (I'm guessing closer to 2017 but we'll see) And we need a couple of years of building/using/repairing them before we see wide spread design wins, then another year before 'memristor' enabled devices hit the market and are or are not competitive.
That said, I'm rooting for them to be successful, flash is in a bad way at the moment with feature size being a hard limit on cell lifetimes.