Read-write device offers new architecture for information processing

Feb 11, 2011 By Miranda Marquit feature

(PhysOrg.com) -- "Silicon based logic devices may run out of steam soon because as devices get smaller, they run into different problems," Laurens Molenkamp tells PhysOrg.com. Molenkamp is a physics professor at Universität Würzburg in Würzburg, Germany. Along with a team of scientists, Molenkamp has been studying ways to make use of different materials and structures to improve logic devices, such as computers.

"Right now, information in computers has to be transferred between logic and memory," Molenkamp continues. "But with memory becoming so big, the process is becoming cumbersome." In order to remedy this problem, Molenkamp and his colleagues in Würzburg have developed a device that allows for memory storage and logic processing in the same structure. A description of their device can be found in : "Fully Electrical Read-Write Device Out of a Ferromagnetic Semiconductor."

Traditionally, information processing is based on different components. Metallic ferromagnets can be used to store information in a remanent manner, such as in a hard disk. Semiconductors are used for logic functions and for volatile memory (RAM). There must communication between memory and logic in order to get the type of computing we are used to. However, there are limitations to this. “ is a problem,” Molenkamp points out. “Additionally, the communication takes time and an enormous amount of interconnects, and there is only so much that can be done when logic and memory are separated in information processing architecture.”

The solution, then, is to create a new information processing architecture that puts logic and memory in the same device. “We have a sample device that we have shown works as a read-write device, putting logic and memory together to create the basis for a new information processing architecture,” Molenkamp says.

In order to create the device, Molenkamp and his fellows used the ferromagnetic semiconductor (Ga, Mn)As. “Our device allows you to perform logic operation with the same circuits where you store info,” he explains. “You can do away with the transfer between logic and memory parts.” This would cut down on heat dissipation, as well as making information processing much faster.

So far, the team at Würzburg has created a one bit device. “There is a little disc in the middle of the device which is the logical bit,” Molenkamp says. “However, in order for our design to be a full logic device, to actually make it programmable, we need two discs touching on each other.” This is what the group is working on now.

In order to take the device further, Molenkamp says that a different set up might be needed. “We were able to show that we could use this device. It is more of a principle of operation,” he points out. “Next, we will have to transfer to a different material that is magnetic at room temperature. We think that our new information processing architecture can carry over to metals.” In order to accomplish this, Molenkamp continues, “one needs to grow crystalline metal layers to use as starting material.” Once that is done, it is possible to begin developing devices that can operate at room temperature, as well as more advanced circuits.

“The adoption of our device could lead to much smaller computers,” Molenkamp says. “Because the type of memory we describe stays encoded, you wouldn’t need RAM, and that would help with heat dissipation and size. We hope that, now that we have shown that you can integrate and logic in this new information processing architecture, that there will be interest in creating devices that use this technology.”

Explore further: First in-situ images of void collapse in explosives

More information: S. Mark, P. Dürrenfeld, K. Pappert, L. Ebel, K. Brunner, C. Gould, and L.W. Molenkamp, “Fully Electrical Read-Write Device Out of a Ferromagnetic Semiconductor,” Physical Review Letters (2011). Available online: link.aps.org/doi/10.1103/PhysRevLett.106.057204

4.6 /5 (23 votes)

Related Stories

New device may revolutionize computer memory

Jan 20, 2011

(PhysOrg.com) -- Researchers from North Carolina State University have developed a new device that represents a significant advance for computer memory, making large-scale "server farms" more energy efficient and allowing ...

DARPA grant to fund research into magnetic logic

Nov 17, 2010

(PhysOrg.com) -- The Defense Advanced Research Projects Agency (DARPA) has awarded a team of researchers led by Wolfgang Porod, Freimann Professor of Electrical Engineering and director of the University of Notre Dame’s ...

A milestone for molecular beams

Oct 29, 2010

(PhysOrg.com) -- When we think of molecular collisions, we often consider massive colliders, like the LHC, sending particles smashing into each other at very high energies. While this is interesting work, ...

Recommended for you

First in-situ images of void collapse in explosives

Jul 25, 2014

While creating the first-ever images of explosives using an x-ray free electron laser in California, Los Alamos researchers and collaborators demonstrated a crucial diagnostic for studying how voids affect ...

New approach to form non-equilibrium structures

Jul 24, 2014

Although most natural and synthetic processes prefer to settle into equilibrium—a state of unchanging balance without potential or energy—it is within the realm of non-equilibrium conditions where new possibilities lie. ...

Nike krypton laser achieves spot in Guinness World Records

Jul 24, 2014

A set of experiments conducted on the Nike krypton fluoride (KrF) laser at the U.S. Naval Research Laboratory (NRL) nearly five years ago has, at long last, earned the coveted Guinness World Records title for achieving "Highest ...

User comments : 18

Adjust slider to filter visible comments by rank

Display comments: newest first

includao
5 / 5 (7) Feb 11, 2011
Such kind of computational architecture is present on neurons.
Quantum_Conundrum
4 / 5 (4) Feb 11, 2011
Such kind of computational architecture is present on neurons.


Yes. While they don't seem to be describing an actual neural net computer, it is a bit similar.

I think this is a step in the right direction. IN the past, I had suggested that we needed to replace sticks of RAM with "processor cards" which would work sort of like Video Cards work now.

This technology seems to take that idea to the ultimate extreme, integrating RAM and processor not only onto the same card or chip, but even the exact same circuits.

If successful, this technology should eventually allow nearly perfect parallel processing. We might even need to re-define what we consider to be a "processor".
Bonkers
5 / 5 (1) Feb 11, 2011
I don't think its quite as revolutionary as it suggests, it is already possible and even mainstream to combine memory and logic - FPGA basic cells are a logic block and a memory cell. CPU architectures are merely logic and adjacent registers, not to mention the on-chip L1 and L2 caches - turns out it is more efficient to rack up the memory cells into arrays, rather than sprinkle logic between them. Even FPGA's with their close logic+cell architecture still use arrayed memory blocks.
Quantum_Conundrum
not rated yet Feb 11, 2011
I don't think its quite as revolutionary as it suggests, it is already possible and even mainstream to combine memory and logic - FPGA basic cells are a logic block and a memory cell. CPU architectures are merely logic and adjacent registers, not to mention the on-chip L1 and L2 caches - turns out it is more efficient to rack up the memory cells into arrays, rather than sprinkle logic between them. Even FPGA's with their close logic+cell architecture still use arrayed memory blocks.


Well, the big difference is a system like this can probably re-configure itself to be optimized for the task at hand.

Maybe we don't need as much memory, but need more processors, it could reconfigure itself to do that, without human input. Maybe some other time we need fewer processors and more memory.

Additionally, such a computer could even be configured to work with variable word size customized to the application, which would be very efficient for some routines.
that_guy
1 / 5 (1) Feb 11, 2011
On a typical high end processor, if there was a 1:1 ratio of transistor to memory bit, that would be about 150 Megabytes for 1.2 billion transistors.

Now, I don't want to be negative or anything, but while having 150MB of on die memory is better than 2(L2 Cache for example), It seems like it would be expensive to create this kind of on die memory...especially enough to replace traditional storage such as a hard drive. it may help alleviate some bottlenecks, but it is almost certainly unable to eliminate the current architecture in use. I call a fail on the extrapolated possibilities of this method.
Quantum_Conundrum
3.3 / 5 (3) Feb 11, 2011
that_guy:

This is not intended to replace hard drives, but RAM and cache.

A "hard drive" is referred to as secondary storage.

"RAM" and "Cache" are primary storage, or "memory".

When someone in the tech world refers to "memory" they are usually referring to RAM or Cache, unless othewise specified. Then you would hear "hard drive" or "ROM". Although technically hard drives are not "RoM" and neither are disk drives any more, but oh well...
that_guy
not rated yet Feb 11, 2011
quantum - the arguments seem to apply in any scenario, although on second read, you may be right that they didn't strongly imply replacing the HD.

The numbers I posted above show that this type of strategy wouldn't work for ram for the same reasons. The added complexity and engineering just to get some extra cache seems pretty futile, given that we already have commercial chips available with 50-100 MB of cache, though they're not common.

feel free to do the math on the numbers, 1 megabyte = approximately 8 million bits. Each transister would handle one bit...

Memrister technology on the other hand could more easily be integrated with todays designs, and doesn't rely on whole new ways of engineering transistors.
El_Nose
1 / 5 (1) Feb 11, 2011
issues with others earlier posts -- it is not true that there is a 1:1 ratio of transistors to bits stored -- it would be lovely if this were true but no memory architecture is that good but 2:1 is about right

secondly its been proven that with the current way memory pages work there is no noticable increase in performace by increase the size of the lowest level L cache beyond a few MB's because 90% of programs simply do not need a tremendous amount of data or instructions. This was an excerise we had to work through in undergrad using pure math to prove it to ourselves. This is very similiar to the fact that throughing cores at a problem does not lead to a linear decrease in computation time at beast 2 cores bring an improvement of 70% not 100% -- in other words 2 cores won;t make it twice as fast - reducing time in half - but it may reduce time by 40%
eric96
1 / 5 (3) Feb 11, 2011
hearing your comments is an interesting read and makes me laugh. Basically, there is no need to put memory on cpu because cpu is not fast enough for this discovery to matter nor is its growth rate; moore's law. Worst, everything is still slowed down by the slowest component; the hard drive. This research is worthless; you will see none of it in computers 20 years from now. Purely theoretical nothing practical. We're not moving to graphene either since that material has too many problems. Rest assure cpu's won't get fast enough even in a decade for this so called revolution to worth any attention cough cough. Silly article for weak minds. I have a lot of these physorg article that come without figures from which necessary conclusions can be drawn. If you can't show us the math, then don't waist our time with your Obama; We Will Change America, Yes We Can

haha-ha--hahahahah
eric96
1 / 5 (2) Feb 11, 2011
hate*
Quantum_Conundrum
2.3 / 5 (3) Feb 12, 2011
eric96:

Actually you're wrong...

The more RAM and Cache a computer has, the less time it wastes doing paging and other querying of the HD.

Don't you remember several years ago during the pentium 1 through 4 era that you had like a gigabyte of "paging" on the HD which was used as "simulated RAM"?

One of the biggest reasons newer computers are exponentially faster is directly related to the fact they have so much cache and RAM and this extremely time consuming process is not longer required.
eric96
1 / 5 (1) Feb 13, 2011
I did say more ram is better. Learn to read.
poof
5 / 5 (1) Feb 13, 2011
Your all wrong, current processors are already fast enough for most tasks, the problem is software is still inefficient and in a constant state of disrepair. Scientists need to be working on redesigning the computer language and how programs are built and maintained before building more hardware we cant use.
ScientistAmauterEnthusiast
1 / 5 (1) Feb 13, 2011
Your all wrong, current processors are already fast enough for most tasks, the problem is software is still inefficient and in a constant state of disrepair. Scientists need to be working on redesigning the computer language and how programs are built and maintained before building more hardware we cant use.


Pretty sure that both are being looked into at the same time, not like such a massive industry isn't multitasking.
Garfieldo
5 / 5 (3) Feb 14, 2011
ok.. took 10sec to register to be able to comment :

I am quite familiar with these works on magnetoelectrics.
1- the concepts in this paper are not new. (check Chiba et al)
2- if you read the paper carefully, you find out that the measurements are done at.... 4.2 Kelvin.. yeah...that's going to be convenient in your iphone...
Ricochet
not rated yet Feb 15, 2011
Further on they say that they feel their methods should be able to be applied to metals at room temperature. That would be a bit more convenient on your Smartphone...
Noidea
not rated yet Feb 17, 2011
Sounds interesting, it could lead to a Russian doll effect, where you could use a mini computer to do work on and if you need more processing power move the files to a larger one.

Or have usb-key that can do processing. That way I could put, lets say a large a large zip file that needs extracting on the usb key take it out and let it do it's work independently so I don't have a performance hit on my bigger machine.

I apologize in advance if it seems like I'm trivialising what could be a major breakthrough for the future of computers.
ECOnservative
not rated yet Feb 18, 2011
Seems similar to the I-RAM project (iram.cs.berkeley.edu) in that it seeks to merge CPU and memory into the same structure. Novel materials, though.