HP Backs Memristor Mass Production

neo12 writes with news that Hewlett-Packard is teaming with Hynix Semiconductor, the world's second-largest producer of memory chips, to mass produce memristors for the first time. Quoting the BBC: "HP says the first memristors should be widely available in about three years. The devices started as a theoretical prediction in 1971 but HP's demonstration and publication of a real working device has put them on a possible roadmap to replace memory chips or even hard drives. ... Steve Furber, professor of computer engineering at the University of Manchester, explained that the potential benefits lie in the fact that memristors are 'much simpler in principle than transistors. Because they are formed as a film between two wires, they don't have to be implanted into the silicon surface — as do transistors, which form the storage locations in Flash — so they could be built in layers in 3D,' he told BBC News. 'Of course, the devil is in the detail, and I don't think the manufacturing challenges have been fully exposed yet.'"

Comments

Haven't heard of this one

[Pojut]

Can any Slashdotters chime in on this...is this technology really "the next thing", or is it "the next thing that is actually nothing"?

Re:Haven't heard of this one

[OzPeter]

Up until recently the memristor was a theoretical fundamental circuit element (like resistor, capacitor and inductor - but the are easy to create in the real world). A few years ago they were actually created and there is a lot of interesting things you can do with the technology

For more info see The mysterious memristor

Or just search Google memristor ieee

Re:Haven't heard of this one

[Anonymous Coward]

Ill let wiki do it for me
http://en.wikipedia.org/wiki/Memristor
http://en.wikipedia.org/wiki/Memristor#Potential_applications

Think flash drives with the access times of DRAM. Think of instant hibernate computers. Instead of having to write out to HD the memory could just remember where it was at. Another application they talk about is crossbar switches. Currently each node in a cross bar has some sort of 'memory' associated to it. Such that the crossbar can 'learn' what are good routes and bad ones. This would allow them to make crossbar switches much smaller and use less power. Crossbar switches are used many times with NxM sized computers and in large communication networks.

They have known about them for ages (since 1971). However, they have only recently figured out how to actually make them at micron sizes.

Now given that we have not seen what they can do. What speeds we are talking about ect... If they can make them however (and at current gate sizes, and volume) the NAND flash drives we use today will probably quickly become a niche product.

Typical application would be somewhere where you want to retain some sort of 'memory' of what is going on but do not want a processor involved. It has also been theorized you could use them for storage of n bits per resistor. So instead of 1 bit per location you could have 4 or even 16 bits. They are also nice in that supposedly you do not need to refresh them as often as we do now so they could also save power.

Yes these things are cool :)

Re:Haven't heard of this one

[Abcd1234]

They have known about them for ages (since 1971). However, they have only recently figured out how to actually make them at micron sizes.

Well, no, not quite. The effect was postulated decades ago, but it was purely theoretical at the time (well, okay, it has been emulated using complicated circuitry). Furthermore, it's not that scientists "figured out how to actually make them at micron sizes"... it's that the effect only comes to the fore at micron sizes, which is why it hadn't been discovered sooner.

No, the discovery is the ability to build a very simple implementation of this theorized circuit element, and its a mighty cool discovery indeed (someone linked to a IEEE article on memresistors... check it out, it's a great read, and does a very good job of explaining the theory and mechanism behind the operation of this particular implementation).

Re:Haven't heard of this one

[mlts]

Memristor based crossbar switches will be extremely useful for two uses:

Shuffling data between VMs in a secure manner on a host such as an IBM 795 or a zSeries that has a large number of VMs in use for different tasks. This way a bunch of VMs that talk amongst themselves frequently (a DB server to an app server) will end up being able to do high I/O without that slamming the CPU.

Another use is tiered memory, where one has a machine with fast RAM and slow RAM, with slow RAM being exponentially faster than going to SSD or magnetic platters. If memristors become able to be printed on a large scale, perhaps we will see machines with 16-32 GB of DRAM, then 256-512 GB of memristive RAM that is used as both swap space, but also a persistant cache for the OS to boot from an image with, never touching the storage media until the OS is fully loaded and the user wants to load documents, or the OS is doing a backup.

Re:Haven't heard of this one

[A nonymous Coward]

You do not RC. MRAM is magnetic, nothing like memresistors.

Re:Haven't heard of this one

[ColdWetDog]

Nature tries to find the most efficient path

Perhaps for particularly weird values of "efficient". If you look carefully at 'Nature' you see kludges upon kludges upon kludges. That's why molecular biology of organisms has been so hard to decipher. Natural selection has had millions of years to twiddle with things - therefore, if something works just a bit better (in terms of organism survival or reproduction) then it can be selected for. It may be a complex, error filled process that slows ten other processes down but given the enormous times that nature gets to play with it doesn't have to be elegant or efficient.

It just has to work a teeny bit better than before.

And of course, that doesn't even begin to talk about the anthropomorphic values that are associated with those adjectives.

Millions of years is a very, very long time.

Can someone explain how the memristor work?

[SharpFang]

Not from a physicist point of view, but from engineer's.

I mean, I know it can store data by means of variable resistance. But how do you read and write? Specific voltages, currents, frequencies? If I understand correctly, it has only two terminals like a resistor. You just apply some variable voltage and measure the current. So how can one differentiate between a write and a read?

Re:Can someone explain how the memristor work?

[Anonymous Coward]

The resistance of the memristor can be viewed as a function of the sum of the current that has passed through it (including the effects of polarity). A write would be performed by sending a larger current through the memristor in the right direction to increase or decrease the resistance; a read would be performed at lower currents that would not change the overall state of the memristor.

Re:Can someone explain how the memristor work?

[42forty-two42]

As I understand it, you apply high-frequency AC to do a read (the current gives you your value). The high-frequency AC ensures that any perturbations to the value cancel out.

Re:Can someone explain how the memristor work?

[Even on Slashdot FOE]

Just make sure to unplug the machine when not in use then, as it will probably be more sensitive to power fluctuations, since they tend to be DC in nature.

Re:Can someone explain how the memristor work?

[42forty-two42]

Presumably they'd have some capacitors over the power supply for the memristor to stabilize the power a bit. And I'm sure they don't use raw AC - there's probably a low-voltage DC inverter in there.

Re:Can someone explain how the memristor work?

[plcurechax]

I actually happened to read this article on IEEE Spectrum about new RAM technologies, and it covers both Phase-Change RAM (PC-RAM), which may of hit a road block in its development, and Resistance RAM (RRAM), of which memristor is a particular kind of.

Re:Can someone explain how the memristor work?

[SharpFang]

I guess they would even contain refresh mechanisms similar to DRAM, run some current in the right direction from time to time.

Re:Can someone explain how the memristor work?

[demonbug]

Presumably they'd have some capacitors over the power supply for the memristor to stabilize the power a bit. And I'm sure they don't use raw AC - there's probably a low-voltage DC inverter in there.

This is HP we're talking about; of course you won't be using raw AC power. These will obviously only work when used in conjunction with genuine HP power supplies; HP can't guarantee that electrical current from other sources won't damage your memristor, so a special chip in the memristor package will ensure that only genuine HP electrical current is used.

Now quit mixing your technical discussion up with my (poor) jokes.

Re:Can someone explain how the memristor work?

[molecular]

as I understand from watching "Memristor and Memristive Systems Symposium" on youtube, low voltage/current (read) wont change the memory. Only applying large voltage changes the state of the memristor. No need to refresh.

layered in 3 dimensions...hmmm

[voss]

3 dimensional memory cubes would be revolutionary, memory sizes could expand geometrically...(only a mild pun intended).

Also with 3 dimensions of wear leveling you probably wouldnt wear them out for a very LONG time.

Re:layered in 3 dimensions...hmmm

[Anonymous Coward]

"memory sizes could expand geometrically"

Yes yes yes, and about 12 picoseconds behind that the software morons will increase bloat exponentially.

Re:layered in 3 dimensions...hmmm

[PPalmgren]

Aren't processors already layered, and we use multiple platters in HDDs? As interesting as a "memory cube" sounds, I would expect heat dissipation and magnetic fields from the current could be a major roadblock in its production.

However, Even if they could only do three layers with insulation layers inserted in between them, it'd probably still be more cost effective than current implementations as long as the memory density is similar.

Re:layered in 3 dimensions...hmmm

[plcurechax]

Aren't processors already layered[?]

None of the in production processors are 3-D layered. Of course all physical devices are three dimensional themselves, but the height is basically moot until designers can use 3-D layering and routing (of the "wires" or interconnections.

I believe research prototypes in universities and R&D labs have been created or are being worked on, so it is not science fiction.

Re:layered in 3 dimensions...hmmm

[Myopic]

You choose which software to run. If you run software that you think is bloated, and everyone else thinks is swell, then you have only violated your own preferences. There is nothing wrong with the software, it's all in your head. I like my 2-gigabyte operating system much better than the 2-megabyte operating system I used in 1993, and I imagine I would like a 2-terabyte operating system very much. I bet it would do all sorts of awesome stuff, and have the kinds of visual effects that I think enhance the graphical experience. If you disagree, that's fine. Version 1.1 of Linux will still run as well as the day it was released, and you can use it.

As for me, I say bring on the bloat. I can hardly wait.

Re:layered in 3 dimensions...hmmm

[Ogi_UnixNut]

From what I've read about memristor's, they don't wear out like Flash does. They are also massively faster than Flash memory. Think of it as a hybrid of RAM and Flash.

In fact, from what I read I think these devices (if they live up to what people are saying about them) will be able to replace both RAM and Disks/SSD's. Instead we'd just have one set of primary memory where everything happens.

Now we just have to see if they can do what is postulated, and how much it will cost to manufacture. If cheap enough to be worth buying due to their benefits, then they will have a huge effect on computing.

Re:layered in 3 dimensions...hmmm

[jandrese]

We could stack chips today except for the fact that it's impossible to cool the middle layers and the thing would almost instantly melt itself down. I wouldn't be surprised if this new technology ran into exactly the same problem.

Re:layered in 3 dimensions...hmmm

[PitaBred]

Considering that a 1.35GB hard drive cost $1800 in 1993, and a 1.5TB (over 1048x the storage) is only $90 right now, and as 2MB is ~.15% of the 1.35GB disk, and 2GB is ~.13% of the 1.5TB disk, I'd say he could like it less and it'd still be an improvement.

Re:layered in 3 dimensions...hmmm

[Myopic]

Okay. So you are looking for, what, Slackware (no bloat)? maybe NetBSD (great security)?

Me, I like pretty icons with fancy animated windows and menus. I like browsers that show flash videos and all kinds of bells and whistles. I love bloat, if that's the kind of thing you mean.

Luckily we both have options. Neither of us is stuck in a position where we want to run software that doesn't exist. The only problem is when someone runs a system that they hate, then complain about it. If you are specifically complaining about Windows (I've never owned a Windows computer, only Macs and Linux), then I will happily join you in panning that turd.

Re:layered in 3 dimensions...hmmm

[Chris Burke]

We could stack chips today except for the fact that it's impossible to cool the middle layers and the thing would almost instantly melt itself down.

Not if it's low enough power. Major chip makers have already considered the minimal level of chip stacking when the "bottom" chip is something relatively low power.

If the promise of using even less power than current flash memory holds true, they may not have any problem stacking chips several layers deep.

Where will the work be done?

[Anonymous Coward]

Seriously, where will the work be done? Will HP set up the fab shop here, or in SK? Or set up multiple shops. I would love to see the DOD suggest to HP that they need to set up a shop here in the USA. We need to make certain that we have our electronics under control here. In addition, the DOD, NSA, etc needs to offer up contracts to American companies that produce equipment here. Why? Because we are increasingly seeing embedded virus, etc coming in from Asia.

Re:Where will the work be done?

[ledow]

Any military that just "relies" on the fact that what the company sends you is actually the device you bought, rather then the one you designed, shouldn't be procuring ANYTHING from ANYONE. If you're buying stuff that was made in some foreign country, and it's for a military application, you should damn well be inspecting it before using it anyway. You should damn well be inspecting it no matter where it came from - even an ally.

And there are a million and one of these "bugs coming from Chinese-manufactured devices" stories and not one shred of evidence that it's ever happened. Hell, the tiniest change to a mass-produced board can have hundreds of subtle knock-on effects in timing, RF interference, capacitance, etc. of the circuit and would most probably break it without some seriously skilled understanding of every tolerance on the circuit - something barely the designers can claim to have. Of course it's *possible*, but it's incredibly, incredibly unlikely and if such militaries were that worried about it, they would, should, can and will make their *own* devices - because it would be no more difficult for a skilled engineer working in a US semiconductor company to affect a circuit than it would be in a foreign semiconductor company.

And incorporating such changes into a foreign military (or even large scale civil) design could actually be perceived as an act of war. The US, EU etc. are NOT at war with anyone at the moment - people need to get this into their thick heads. Starting a war because you want to listen to a couple of phone calls is not a sensible way to act when one side has the world's largest nuclear arsenal, or the world's largest airforce, or navy, or spy satellite system.

Please stop the "foreigners are bad" crap. That's what started the last "war" the US had.

Re:Where will the work be done?

[smellsofbikes]

Seriously, where will the work be done? Will HP set up the fab shop here, or in SK? Or set up multiple shops. I would love to see the DOD suggest to HP that they need to set up a shop here in the USA. We need to make certain that we have our electronics under control here. In addition, the DOD, NSA, etc needs to offer up contracts to American companies that produce equipment here. Why? Because we are increasingly seeing embedded virus, etc coming in from Asia.

Last I heard, Hewlett Packard still has domestic fabs: Corvallis, Oregon, I believe, and I think they still have fab capability in Palo Alto. A lot of large semiconductor companies retain small cutting-edge fabs at their headquarters for doing small runs of experimental stuff.

My company has two of its fabs in the US and one in the UK. We do all our packaging and testing in southeast Asia, but the company apparently made a decision about three years ago that we're not going to put fabs there, because we closed down one we'd built less than two years before.

And, as I've said many times before, with the possible exception of processors, it's really difficult to sabotage chip design. Your profit margin is directly related to the surface area your chip layout occupies, so it is aggressively minimized in design, and there simply isn't room on the silicon to splice in new stuff. Added to that, chip companies that I've worked with usually do a planet run of many prototype silicon designs through one fab, often their domestic/in-house fab, do their initial testing on that, and only after that do they put it into production with a full-size mask on dedicated silicon in the production fab, so if you wanted to sneak stuff in you'd have to infiltrate both fabs, or you'd end up with silicon that's visually different -- and we spend a lot of time with high-power microscopes and microprobes poking around at new silicon, sometimes chipping bits out with a laser if we need to do a metal layer change, so it's not like someone wouldn't notice changes on smaller chips. And even if all of THAT didn't catch changes, test and product engineers spend months writing automated test programs that check each pin on each chip and characterize its leakage current, its current draw when functioning, its ESD resistance, all sorts of things, and added circuitry will change those values.

If a company doesn't have a fab, and they just send all their completed masks (or, even worse, just the designs) off to one company, then I think it's possible, albeit difficult, for stuff to sneak into the silicon. But a company that has a fab, or runs their designs through multiple fabs, is pretty unlikely to get compromised silicon without noticing it. It would be significantly easier for a malevolent group to just design their own silicon from the ground up, and package it to look like the target chip and get it into distribution channels by selling it as authentic stuff, than to try to compromise a company's silicon.

the math has more to it

[Anonymous Coward]

among other things, memristive slime mold modeling and synaptic chain behaviours...

Re:Neural nets

[MadKeithV]

One step closer to real AI?

As opposed to artificial AI?

Just big chips?

[shaitand]

I can buy big bags of transistors direct from hong kong on ebay for $5 for my hobby electronics use. Will that ever be the came with the memristor or will these never be made in component size and instead restricted to larger chips that are tens of dollars or more capacities only?

Re:Just big chips?

[Amouth]

considering that the effect is only exploitable at nano meter scales - it might take a lot of work/time before anyone makes them into something you can buy off the shelf as a single component.

Re:Just big chips?

[iprefermuffins]

Yo dawg, I heard you like regular expressions...

Steve Furber

[Alioth]

In case you're wondering who Steve Furber is, he is also one of the original designers of the ARM CPU.