EEVblog Electronics Community Forum
General => General Technical Chat => Topic started by: nixfu on December 16, 2014, 06:26:13 am
-
Interesting article about what HP thinks will soon be replacing DRAM technology, static ram, flash, and hard drives, and pretty much the entire computer architecture as we know it. HP seems to think it will destroy the entire current way we do computing with processors, memory, cache, storage, systems today.
http://highscalability.com/blog/2010/5/5/how-will-memristors-change-everything.html (http://highscalability.com/blog/2010/5/5/how-will-memristors-change-everything.html)
"Memristors are nano devices that remember information permanently, switch in nanoseconds, are super dense, and power efficient. That makes memristors potential replacements for DRAM, flash, and disk."
"How much storage are we talking about on a single chip? With an invention they hail as important as the memristor is a new architecture that allows the stacking of multiple crossbar memories on top of each other. This allows multiple petabits of memory (1 petabit = 128TB) to be addressed in one square centimeter of space."
" But wait, there's more! Memristors are not just stuck in they past, they don't just remember, they can perform logic! I find this completely strange. We're not used to our memory also acting like a CPU. But it turns out memristors naturally implement something called material implication logic, which can be interconnected to create any logical operation, much the same way NAND gates were used to build early supercomputers because they were easier to build."
-
4 and a half years and still the same thing, did anything come out of this memristor thing?
BTW there are many posts about memristors in the forum but I think this was the first on in this forum:
https://www.eevblog.com/forum/chat/memristor-what/ (https://www.eevblog.com/forum/chat/memristor-what/)
So it's been quite a while since this groundbreaking tech, I guess they gave up and now they are focusing in 3d printers.
-
From what I've heard memristors are really great but like graphene they're having real difficulty making production quantities.
-
This has been "The next big thing" since the 1980's, like bubble memories.
-
I'd like to see some at digikey (or an equivalent distributor) before I buy into the hype too much. Even if they are $50 for a single 0603 package due to low yield, at least that would get the ball rolling. My main argument: which came first, the capacitor or DRAM?
But I'm guessing that isn't going to happen, because HP probably holds all of the patents, and they aren't going to release any devices until they have perfectly working memristor-based RAM (aka a long time away).
-
If I had a dollar for every new idea that was going to make DRAM or flash obsolete, I could afford to finance some serious memory research. ;)
-
The logic BTW is essentially the charge mode analog of saturable / magnetic logic, which depends on integral voltage to control current (nonlinear inductance). Memristors depend on integral voltage to control integral current, and (as far as I know) is similarly nonlinear in most or all instances.
Tim
-
4 and a half years and still the same thing, did anything come out of this memristor thing?
BTW there are many posts about memristors in the forum but I think this was the first on in this forum:
https://www.eevblog.com/forum/chat/memristor-what/ (https://www.eevblog.com/forum/chat/memristor-what/)
So it's been quite a while since this groundbreaking tech, I guess they gave up and now they are focusing in 3d printers.
First of all, it is not the same thing as 4 years ago. Secondly, how much time do you think it takes to develop breakthrough technology to make it commercially viable to push into production?
Memristor research at HP Labs (https://www.youtube.com/watch?v=7inxPhbe-EI#ws)
-
Wow, HP uses a lot of non-agilent gear!
-
First of all, it is not the same thing as 4 years ago. Secondly, how much time do you think it takes to develop breakthrough technology to make it commercially viable to push into production?
Up to the end of the universe in which case it wasn't breakthrough or commercially viable.
There seems to be a common stupidity that anything is possible given enough time and money. Memristors may be breakthrough and commercially viable, but, until it has actually been done we don't know.
-
First of all, it is not the same thing as 4 years ago. Secondly, how much time do you think it takes to develop breakthrough technology to make it commercially viable to push into production?
Up to the end of the universe in which case it wasn't breakthrough or commercially viable.
There seems to be a common stupidity that anything is possible given enough time and money. Memristors may be breakthrough and commercially viable, but, until it has actually been done we don't know.
Reminds me about my friend who asked why they cannot just make computers orders of magnitude better right now and technology is developing so slow. My answer was - "Are you fucking stupid? Just look what we had 2 decades ago. And 50 years ago nobody could even dream about it." It just fascinates me how people expect that everything should appear right now and from nowhere. Just think about blue/white(blue+phosphor) LEDs. How much time it took to develop blue LEDs since first LEDs appeared. And how much time passed since blue LEDs appeared for LEDs to become viable for lightning applications.
-
Remeber, HP is a computer company, now. They don't want their PC business, so you can consider them an IT server company.
That's where the first uses of memristors will be.
http://arstechnica.com/information-technology/2014/06/hp-plans-to-launch-memristor-silicon-photonic-computer-within-the-decade/ (http://arstechnica.com/information-technology/2014/06/hp-plans-to-launch-memristor-silicon-photonic-computer-within-the-decade/)
-
oh fuff. they've been blabbing about this and phase change memory for so long now ... i still have to see the first working devices.
sure they can make a few kilobytes of this stuff. but the gigabyte devices ? nah. never gonna happen
-
Whatever happened to racetrack memory? Is that still a thing?
-
I guess the big development is that HP says in the next year they will actually be releasing a radically new computer platform based on memristors.
http://www.extremetech.com/extreme/196003-hp-reveals-more-details-about-the-machine-linux-os-coming-2015-prototype-in-2016 (http://www.extremetech.com/extreme/196003-hp-reveals-more-details-about-the-machine-linux-os-coming-2015-prototype-in-2016)
"HP has been working on a radically new type of computer, enigmatically called The Machine (not this machine). The Machine is perhaps the largest R&D project in the history of HP. It’s a complete rebuild of both hardware and software from the ground up. A massive effort. HP hopes to have a small version of their datacenter scale product up and running in two years."
http://highscalability.com/blog/2014/12/16/the-machine-hps-new-memristor-based-datacenter-scale-compute.html (http://highscalability.com/blog/2014/12/16/the-machine-hps-new-memristor-based-datacenter-scale-compute.html)
HP makes it sound like this is some sort of revolutionary bet-the-company r&d effort. But, frankly I believe that NOTHING innovative can ever come out of a corporation. They design everything by consensus, take zero risks, and approve everything by committee. The only thing you can ever get approved in that world is incremental small changes to existing products. Real innovation happens outside of that sort of world and that is why large companies usually get all their innovation by buying smaller companies.
-
HP makes it sound like this is some sort of revolutionary bet-the-company r&d effort.
More likely HPLabs is under threat, and this is their "I can teach a horse to sing in five years" response.
But, frankly I believe that NOTHING innovative can ever come out of a corporation. They design everything by consensus, take zero risks, and approve everything by committee. The only thing you can ever get approved in that world is incremental small changes to existing products. Real innovation happens outside of that sort of world and that is why large companies usually get all their innovation by buying smaller companies.
To use the technical term, bollocks.
Especially where HP and HP Labs were concerned. Go and do a cursory examination of their history and achievements.
-
I believe that NOTHING innovative can ever come out of a corporation. They design everything by consensus, take zero risks, and approve everything by committee.
That is a lot less true than you'd think. It certainly is true in a lot of places, but I've worked for a couple big places that are very, very pro-innovation internally, in words and in action.
Currently, I work for a large, very well known corporation, and I'm in an area that has about a dozen and a half people working on skunkworks-type projects full-time. Historically, we have been given carte blanche with spending when it comes to innovation projects. There are other areas doing similar work. We are basically left alone to innovate, failing or succeeding based on the merits of our ideas and efforts.
-
Especially where HP and HP Labs were concerned. Go and do a cursory examination of their history and achievements.
Sure when like Bill or Dave was still around at HP, or Jobs was still at apple, but after those visionaries leave and you have the Harvard MBA CEO's take over, it pretty much spells the end of innovation.
-
It sounds like an extremely interesting possibility... of course we'll see how fast they can adopt it.
Let's not forget that new computation paradigms need to be figured out and adopted. Ie. how can you make use of a memory that can do logic? Can you apply it meaningfully in a Neumann architecture?
If you can't, chances are it will never catch on outside speciality applications. There have been so many groundbreaking innovations that would mean moving to a non-Neumannian architecture... and none of them are in particularly widespread use in general purpose computing. Because most people can't code for them, that's why. :)
Also, as regular memory, there's the question of how it would compare to DRAM in terms of functionality vs. cost. DRAM is pretty good for what we use it for. How would price for speed and size compare?
There are lots of questions with all new technologies.
-
Especially where HP and HP Labs were concerned. Go and do a cursory examination of their history and achievements.
Sure when like Bill or Dave was still around at HP, or Jobs was still at apple, but after those visionaries leave and you have the Harvard MBA CEO's take over, it pretty much spells the end of innovation.
Which is an entirely different point to your initial contention, viz "But, frankly I believe that NOTHING innovative can ever come out of a corporation. They design everything by consensus, take zero risks, and approve everything by committee."
Now all you have to do is explain Bell Labs and Xerox PARC achievements. And, as someone that has knowledge of electronics and software, I'm sure you are already aware of their contributions without having to do any research whatsoever.
Summary: while I have a similarly low opinion of MBAs, your statement is simply wrong.
-
Summary: while I have a similarly low opinion of MBAs, your statement is simply wrong.
Agreed. Anyone that takes their MBA seriously prefers their brain to be in stupid-mode, and no amount of evidence will get a true MBA-believer to ever change how they run things.
And, like a lot of folks, I've had plenty of authority figures screw me over handily just because they could.
The statement that no innovation can come from a large organization is just not based on fact. It may be based on someone's personal experience, and I'm absolutely sure that there are people who have only worked at large organizations where innovation is viewed as risk and the MBA mantra "avoid risk at any cost" kills innovation quickly. Not everywhere is like that, though.
The wider scope some of us can see just makes it a clearly false statement.
Nixfu, if you're young, you'll wind up working for a place that values innovation at some point. Keep your eyes out for it, because if you go in with the assumption that it won't happen, then there's no way you'll ever see it, even if it's right in front of you.
-
Let's not forget that new computation paradigms need to be figured out and adopted. Ie. how can you make use of a memory that can do logic? Can you apply it meaningfully in a Neumann architecture?
HP has been talking about Bertrand Russell for the last four years. They went balls deep into the idea of building mainframes based on new computation model with this thing. They sure as hell wont let memristor become commodity = they wont sell it on open market = you can all forget about x86 workstations with memristor storage.
On one hand its pretty cool, if it works it will be as disruptive as anti gravity engine or teleporter. If it doesnt we as a species will waste 10-20 years of progress that could be had by incorporating memristors as simple storage right now.
-
Let's not forget that new computation paradigms need to be figured out and adopted. Ie. how can you make use of a memory that can do logic? Can you apply it meaningfully in a Neumann architecture?
If you can't, chances are it will never catch on outside speciality applications. There have been so many groundbreaking innovations that would mean moving to a non-Neumannian architecture... and none of them are in particularly widespread use in general purpose computing. Because most people can't code for them, that's why. :)
y
New programming paradigms will undoubtedly be required in the future, but the von-Neumann architecture isn't the source of the problem.
The particular pain point for HPC and similar activities is the assumption that all processors can access all memory; this over-stresses the interconnect and is limited by synchronisation/exclusion operations. The solution will come in the form of new was of structuring computation, and that benefit greatly from new high level languages that directly support the computation model.
Which is a tall order that won't come easily.
-
Both PCM and MRAM have been shipped in commercial devices ... although Micron threw in the hat for PCM.
Ferroelectric tunnel junctions look most promising to me, maybe once the patents run out some of that promise can be realized.
-
HP has been talking about Bertrand Russell for the last four years. They went balls deep into the idea of building mainframes based on new computation model with this thing. They sure as hell wont let memristor become commodity = they wont sell it on open market = you can all forget about x86 workstations with memristor storage.
Patents expire, and we'll see memristors in purchasable computers before the patents expire.
The more disruptive a technology is, the longer it takes that technology to be fully utilized in commercial products. Slightly improved LEDs appear on the market nearly as soon as they are invented, because everyone knows how to use an LED. Memristors will take much longer to be understood well enough to be used in all the ways they will eventually be used
HP better get on the ball fast or their patents will be vapor before they are able to license the tech to anyone who can make a useful commercial product with them.
In the meantime, someone should invent a memristor simulator so gates and such can be worked out.
-
Gates are not able to be 'worked out', thats the whole point of mentioning Russell, you need new logic model.
HP will die before letting you use them, they want to monopolize hi end market with memristors, and are not interested at all in cooperating. I think they want to become IBM from the seventies.
-
I don't see how memristor is an enabler for the distributed processor in memory architecture, in fact it's more like an extra hurdle since some of the lithography steps are an even bigger departure than DRAM. Optical interconnects and greater integration seem to be the enabling technology for getting denser Machines, not memristors.
Memristors "merely" make snapshotting easier, which enables things like persistent computing (ala. KeyKOS) and is important for speeding up supercomputers (faults are the rule now, not exceptions).
-
HP will die before letting you use them, they want to monopolize hi end market with memristors, and are not interested at all in cooperating.
Do you have evidence for that assertion, or is it an uninformed opinion?
The reason I ask is that HP's history very definitely includes examples of HP cooperating technically with another company and simultaneously commercial competing with that company.
HP will not fabricate commercial memristors; IIRC they are in bed with Samsung. Ditto Itanium/Intel.
-
Memristors "merely" make snapshotting easier, which enables things like persistent computing (ala. KeyKOS)
It isn't quite that simple.
Personally i wonder how they will avoid the disadvantages of Smalltalk's images - which sound remarkably similar in some respects.