backtop


Print 80 comment(s) - last by HollyDOL.. on Dec 4 at 4:02 PM


The memristor looks unassuming; here's a single titanium dioxide memristor up close.  (Source: J. J. Yang, HP Labs)

HP has created the world's first memristor circuit. Researchers cut out transistors from the bottom layer of this silicon-based chip (shown in yellow and blue) and replaced them with fewer memristors in the top layer (shown in red). The device showcases the power of the memristors.  (Source: Qiangfei Xia, HP )
HP's new integral circuit component allows engineers to produce identical logic circuits using fewer transistors and less space

The goal of chipmakers has always been to push Moore's Law, squeezing more and more transistors into a smaller space.  But what if you could do more with fewer transistors?  That's the intriguing potential of HP's memristor, which joins the standard resistor, the capacitor, and the inductors as a fabled fourth integral circuit component.

First envisioned in 1971 by Berkeley professor Leon Chua, a memristor is a device which can vary its resistance based on the magnitude and direction of the voltage of an applied signal.  Furthermore, it retains its resistance state even if it is powered off.

Rediscovering Professor Chua's groundbreaking, but largely overlooked work, engineers and researchers at HP Labs dug into the problem of creating a memristor on the nanoscale.  In May they finally succeeded, creating the world's first memristor.

This week at the newly created Memristor and Memristor Systems Symposium, in Berkeley, CA the true potential of the unleashed memristor has finally begun to be seen.  One thing is clear -- the little device has the potential to rock the entire hardware industry.

When paired with transistors, memristors can be used to create new and unique circuits that function exactly like circuits with many more transistors.  The new circuits are much smaller and consume far less power.  In short, memristors allow you to do more with less.

Lead researcher Stan Williams, a senior research fellow at HP, states, "We're trying to give Moore's Law a boost."

Indeed, HP's new invention could allow licensed chipmakers to not only continue Moore's law, but to almost instantly leap ahead, shifting Moore's law years ahead.  Williams describes this new mentality, stating, "We're not trying to crowd more transistors onto a chip or into a particular circuit.  Hybrid memristor-transistor chips really have the promise for delivering a lot more performance."

In the past chipmakers have developed circuit elements consisting of multiple transistors to do the job that a single memristor does.  By chopping out these transistors and putting a memristor in their place, the circuit uses less power and is shrunk.  HP has demonstrated such a deployment in the first ever working memristor-transistor hybrid chip

Mr. Williams says making the device was easier than expected.  He states, "Because memristors are made of the same materials used in normal integrated circuits it turns out to be very easy to integrate them with transistors."

Mr. Williams and HP researcher Qiangfei Xia led a team which developed the circuit, a new type of field-programmable gate array (FPGA) which uses far fewer transistors by employing semiconductor titanium dioxide memristors. 

FPGAs are reprogrammable hardware circuits, one of the hottest fields in computer engineering today.  While FPGAs are frequently used by engineers to test their circuit designs on a smaller scale, as they're reconfigurable, they're too expensive, slow, and power-hungry for normal circuits.  Typically they are replaced by leaner dedicated circuits based on the optimized FPGA design.  Mr. Williams continues, "When you decide what logic operation you want to do, you actually flip a bunch of switches and configuration bits in the circuit.  What we're looking at is essentially pulling out all of the configuration bits and all of the transistor switches."

The new memristor-sporting FPGA design is more compact, more affordable, and uses far less power.  In short, it could become the first FPGA to be a viable competitor to dedicated silicon circuits.  The potential is impressive; imagine buying AMD or NVIDIA's latest graphics card and receiving regular hardware updates to increase performance and remove errata.  As Mr. Williams puts it, "If our ideas work out, this type of FPGA will completely change the balance."

Aside from traditional processing circuits, memristors are also very promising for flash memory, and could greatly reduce its cost.

HP researchers say that the biggest obstacle to memristor circuits is the lack of familiarity among engineers with the device.  However Mr. Williams and others at HP assure that the public will see memristor circuits within three years, and that the device has the potential to eventually transform the entire computing industry.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Another 3-5 years...?
By MadMan007 on 11/26/2008 12:51:28 PM , Rating: 4
Let's see if this is another one of the ever present '3-5 years away from application' technologies or not. I like reading about this stuff but it's just geeky cool tech pron until actual real products are announced. I hope HP licenses it and lets others that have huge production facilities run with it.

I don't know about this though...
quote:
imagine buying AMD or NVIDIA's latest graphics card and receiving regular hardware updates to increase performance and remove errata.
I can see AMD maybe doing that but greedyNV? Hardly...they'd rather sell you the same chip repackaged through 3 generations and neutered by driver limitations than update the old one.




RE: Another 3-5 years...?
By SavagePotato on 11/26/2008 12:56:37 PM , Rating: 5
Or release buggy products they can fix later to meet deadlines. RE: my separate post on the subject.

Or ransom performance to you with payed upgrades, I could see that being Nvidia's style.


RE: Another 3-5 years...?
By Lightnix on 11/26/2008 2:49:40 PM , Rating: 2
Then we get back to the whole 'hardware piracy' problem that also comes up in discussion about nano-technology microprocessor fabrication sci-fi style stuff.

I for one would love to be able to buy a geforce 9600 GSO and have it upgraded to 9800 GTX+ just by torrenting some weird chip design patch.


RE: Another 3-5 years...?
By Kamasutra on 12/1/2008 4:09:28 PM , Rating: 2
Or possibly better yet, a chip design patch to get the most out of the specific game you're playing. This could also do wonders for GPGPU computing.


RE: Another 3-5 years...?
By Ringold on 11/26/2008 4:21:59 PM , Rating: 2
First of all, AMD could use some greed and capitalist drive like that seen in Nvidia. If they don't.. they'll cease to exist.

Or maybe not. Maybe we'll bail them out too. Why the hell not at this point?

But moving on.. second of all.. I would think you'd still buy new chips. More of these new kinds of circuits, smaller ones on new manufacturing processes, etc. Performance could thus still be improved, no? Just like the 'tick' in Intel's tick-tock cycle. Or maybe it's the tock. Which ever.

They could also charge, as others have pointed out, for the hardware updates. Given all the money that goes in to design, that would only be fair. Maybe minor patches for a given design could be free, but whole new designs with large gains/advantages.. no way.


RE: Another 3-5 years...?
By foolsgambit11 on 11/26/2008 4:42:44 PM , Rating: 2
How long will HP's patent be on this? Because that's how long it will be before we see it in the majority of applications, I'd say. Why bother licensing memristor technology from them when you can get the job done pretty well with the tools we have now? That's what the hardware developers' bosses will say, at least. The developers would love to begin using these in their designs right away, I'm sure.

I'm betting there's a speed problem with these things. How quickly can it change resistance? That would be the limiting factor, and HP doesn't explicitly say they are fast. It says they can make designs smaller and more power efficient. It says it will giver Moore's Law a boost (by which I assume they mean number of transistors (or transistor-equivalents, I guess) in a given area) - but that doesn't necessarily mean increased speed when you're no longer using transistors. Maybe I'm just a pessimist. Anybody got some data on performance, or is it just too early in development to say how these will perform?

Also, what kind of percentage die shrink are we talking about? I'm assuming it would depend on the application (the number of circuits that can be reproduced with memristors instead of transistors). So what would we expect in usage for CPUs? GPUs? non-volatile memory?


RE: Another 3-5 years...?
By drebo on 11/26/2008 5:19:00 PM , Rating: 2
That's not true at all. If nVidia, for example, can save $10/chip by licensing this for $5/chip, then of course they will go for it. Remember, the cost of a chip in raw materials is directly proportional to its size, rather than the complexity of the circuits inside of it. Anything that can make that chip smaller will save money.


"DailyTech is the best kept secret on the Internet." -- Larry Barber














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki