backtop


Print E-mail del.icio.us 16 comment(s) - last by s12033722.. on Mar 26 at 3:57 PM


Laser waveguide machines were primarily designed as large, bulky lab experiments like this prototype EUV device in 2003. Now Sun claims it can commercialize the technology to make it small enough to put inside a blade server.   (Source: University of Colorado)
A new alignment and waveguide technology under Sun's care may give birth to a new breed of multi-core supercompters

Last December DailyTech reported on IBM's work involving light-based microchip interconnects for multi-core processors. Just this month IBM demonstrated a nanophotonic switch to help control the flow of optical data. IBM seems to be on the fast track to bringing photonic waveguide interconnects to bear.

But IBM isn't the only company working with lasers. Today, Sun Microsystems announced a $44 million, DARPA funded project to research their version of the waveguide interconnect.

The leverage behind photonic microchip interconnects is multi-fold. Not only can more data be transmitted through light than through electricity, the power needed for the same amount of data is many times less, the heat generated is minimal, and the space needed for the physical part of the interconnect is much smaller thanks to not having to deal with things like leakage and heat.

The underlying idea behind Sun's new interconnect is that each chip in an array would be able to communicate directly with any other chip via waveguided lasers. This would eliminate the bottlenecks that multi-processor systems and supercomputers deal with currently when data must be transferred between chips. The new system is expected to be able to move several billion bits of data per second within the array.

“This is a high-risk program. We expect a 50 percent chance of failure, but if we win we can have as much as a thousand times increase in performance,” said Ron Ho, a Sun researcher and co-leader of the interconnect project in a New York Times interview.

Sun's new waveguide interconnect competes directly with IBM's recently announced waveguide alternative.  IBM, Intel and Hewlett-Packard lost the DARPA bid.


However, Sun representatives emphasize the reason they were chosen over the other candidates for the funding revolved around how close the company is to commercializing the technology.  Sun demonstrated its first waveguide interconnect in 2003.

The real feat, a struggle for all the chip manufacturers to date, has been precisely aligning semiconductors that use waveguide interconnects.  So far though, the only real application Sun touts is processor-to-processor transmission.

Sun dubs each group of processors connected via the waveguide interconnect as a "macrochip."  The individual processors inside the macrochip will communicate to each other at such high transmission rates that they will replace a traditional processor node or blade in a high performance setup.

Sun's new macrochip system could pave the way to a new generation of super computers or it could end up a flop. Either way, they still face competition from other groups like IBM, Intel and Hewlett Packard and MIT. One of these systems will likely work and the future of multi-core and multi-CPU interconnects looks very bright indeed.



Comments     Threshold


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

By roadrun777 on 3/25/2008 11:09:39 PM , Rating: 2
It amazes me that the turn around time for R&D is like 20 years for tech like this. I happen to believe that it shouldn't take 20 years for research to make it into mainstream production. If they demonstrated a working model back in 2003, what have they done since then? Shelved it and waited for ... ?

Seems like there is way too much red tape to cut before products make it to market.




By eman7613 on 3/26/2008 12:05:57 AM , Rating: 2
proof of concept can be miles away from practical


By SiN on 3/26/2008 12:18:24 AM , Rating: 2
oh yeah!!! what about SED's...? hmmmm?


By elessar1 on 3/26/2008 9:15:19 AM , Rating: 2
or ligth years away.....

sorry... :P


By elessar1 on 3/26/2008 9:22:01 AM , Rating: 2
or ligth years away.....

sorry... :P


By fezzik1620 on 3/26/2008 8:16:46 AM , Rating: 3
I assume you are not serious. But for all those who read this post and agreed...
The reason why it takes 20 years for something like this to become market ready is because for everything they try that works, there were 20 things that they tried that didn't work. And they may get so far in the design process and then realize that one trail they went down isn't going to work, but they didn't know that until they put months of work into getting something they could test a theory with; then it's back to the drawing board. I'm sure that some work has gone on on this project since 2003, but if progress has slowed, that thing that they've "waited for..." is that $44 million grant from DARPA mentioned in the article. Lasers don't grow on trees, son.

Please don't post just to start a flame war. If you want to mark your place in history write a book or something.


It seems like everyone these days is copying scifi
By Samus on 3/25/2008 11:11:20 PM , Rating: 2
How many scifi's you seen that had computers with lasers in them? I'm thinking of a couple... :)




By Omega215D on 3/26/2008 4:40:47 AM , Rating: 6
SPARCs with frickin' laser beams attached to their heads?


this is the revolutionary, not the evolutionary step
By nerdye on 3/25/2008 11:45:39 PM , Rating: 1
This is revolutionary in the fact that it will change the way that a "bus" transfers data in the most raw sense possible. Electrical "bus" has bean around forever, but this technology poses to replace that and have such data optically transfered.

Sure AMD64 was a great evolutionary development as it put the memory controller on die to the cpu and greatly reduced latency from communication of devices, intel will soon implement this same scheme with their upcoming nahelem, which is also a great evolutionary step for them.

Optical is the future, this is revolutionary and will change the fundamentals of computing. Sun is skeptical saying they may fail or may find success, that is irrelevant in the end, we are to anticipate this technology whether its ibm, intel, amd, or anyone else to first develop it. I applaud you sun, as I look up to the stars, the future is coming.




By webada on 3/26/2008 1:24:30 PM , Rating: 2
i'm sorry but i can't agree that is considered to be evolutionary and certainly does not change the fundamentals of computing (that statement I would reserve to quantum computing that MIT is currently researching)

laser/photons or semicondutors, the fundamentals don't change. Whether is bus or switch fabic, its still reduced to the basis of a turing machine. That fundamental doesn't change, only the delivery mechanism.


By webada on 3/26/2008 1:28:19 PM , Rating: 3
*considered to be revolutionary. Its evolutionary.


By s12033722 on 3/26/2008 3:57:45 PM , Rating: 2
I agree - this is an evolutionary step in high-speed interconnect. Optical interconnect in itself is hardly new, this is simply an application of existing optical interconnect technology to the chip scale. Ethernet (though most is now wired), fiberchannel, optical fiber backbones - these are all examples of media that use similar ideas, but on a more macro level.


Speaking of SciFi
By wrack on 3/25/2008 11:39:30 PM , Rating: 2
I have seen computers using Bio-Neural Gel Pack.




RE: Speaking of SciFi
By webada on 3/26/2008 1:25:30 PM , Rating: 2
that's another form of semiconductors...


punny
By SiN on 3/26/2008 12:17:11 AM , Rating: 2
oohhh... ho ho ho xD

"the future of multi-core and multi-CPU interconnects looks very bright indeed."




indeed
By jeddragoon on 3/26/2008 1:27:40 AM , Rating: 2
..lol very bright future .indeed..but then they have to figure how to built motherboard first right ..maybe fiber optic base..1/10th lightspeed




"Well, we didn't have anyone in line that got shot waiting for our system." -- Nintendo of America Vice President Perrin Kaplan

DailyTech Poll
Which web browser do you use on your primary personal machine? 






44 Comments












botimage
Copyright 2009 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki