IBM Researchers Develop World's Tiniest Nanophotonic Switch to Route Optical Data Between Cores in Future Computer Chips

ericlp

Diamond Member
Dec 24, 2000
6,139
236
106
http://money.cnn.com/news/news...marketwire/0375342.htm

This is really good news! This may be the finial chip evolution design the world has been waiting for...

IBM (NYSE: IBM) scientists today took another significant advance towards sending information inside a computer chip by using light pulses instead of electrons by building the world's tiniest nanophotonic switch with a footprint about 100X smaller than the cross section of a human hair.

The switch is an important building block to control the flow of information inside future chips and can significantly speed up the chip performance while using much less energy.

Today's announcement is a continuation of a series of IBM developments towards an on-chip optical network:

In November 2005, IBM scientists demonstrated a silicon
nanophotonic device that can significantly slow down
and actively control the speed of light.
In December 2006 an analogous tiny silicon device was used
to demonstrate buffering of over a byte of information
encoded in optical pulses a requirement for building
optical buffers for on-chip optical networks.
In December 2007, IBM scientists announced the development
of an ultra-compact silicon electro-optic modulator,
which performs the job of converting electrical signals
into the light pulses, a prerequisite for enabling on-chip
optical communications.

"This new development is a critical addition in the quest to build an on-chip optical network," said Yurii Vlasov, manager of silicon nanophotonics at IBM's T.J. Watson Research Center. "In view of all the progress that this field has seen for the last few years, it looks that our vision for on-chip optical networks is becoming more and more realistic."

Today's announcement is another significant advancement in their quest to develop next-generation high-performance multi-core computer chips which transmit information internally using pulses of light traveling through silicon instead of electrical signals on copper wires.

In a paper published in the journal Nature Photonics, IBM unveils the development of a silicon broadband optical switch, another key component required to enable on-chip optical interconnects. Once the electrical signals have been converted into pulses of light, this switching device performs the key role of "directing traffic" within the network, ensuring that optical messages from one processor core can efficiently get to any of the other cores on the chip.

The IBM team demonstrated that their switch has several critical characteristics which make it ideally suited for on-chip applications. First, the switch is extremely compact. As many as 2000 would fit side-by-side in an area of one square millimeter, easily meeting integration requirements for future multi-core processors.

Second, the device is able to route a huge amount of data since many different wavelengths or "colors" of light can be switched simultaneously. With each wavelength carrying data at up to 40 Gb/s, it is possible to switch an aggregate bandwidth exceeding 1 Tb/s -- a requirement for routing large messages between distant cores. Last but not least, IBM scientists showed for the first time that their optical switch is capable of operating within a realistic on-chip environment, where the temperature of the chip itself can change dramatically in the vicinity of "hot-spots," which move around depending upon the way the processors are functioning at any given moment. The IBM scientists believe this temperature-drift tolerant operation to be one of the most critical requirements for on-chip optical networks.

Holy CRAP!!!

An important trend in the microelectronics industry is to increase the parallelism in computation by multi-threading, by building large scale multi-chip systems and, more recently, by increasing the number of cores on a single chip. For example the IBM Cell processor which powers Sony's PlayStation 3 gaming console consists of nine "brains," or cores, on a single chip. As users continue to demand greater computing performance, chip designers plan to increase this number to tens or even hundreds of cores.

This approach, however, only makes sense if each core can receive and transmit large messages from all other cores on the chip simultaneously. The individual cores located on today's multi-core microprocessors communicate with one another over millions of tiny copper wires. However, this copper wiring would simply use up too much power and be incapable of transmitting the enormous amount of information required to enable massively multi-core processors.

IBM researches are exploring an alternative solution to this problem by connecting cores using pulses of light in an on-chip optical network based on silicon nanophotonic integrated circuits. Like a long-haul fiber-optic network, such an extremely miniature on-chip network will transmit, receive, and route messages between individual cores that are encoded as a pulses of light. It is envisioned that using light instead of wires, as much as 100 times more information can be sent between cores, while using 10 times less power and consequently generating less heat.

The report on this work, titled "High-throughput silicon nanophotonic wavelength-insensitive switch for on-chip optical networks" by Yurii Vlasov, William M. J. Green, and Fengnian Xia of IBM's T.J. Watson Research Center in Yorktown Heights, N.Y., is published in the April 2008 issue of the journal Nature Photonics. This work was partially supported by the Defense Advanced Research Projects Agency (DARPA) through the Defense Sciences Office program "Slowing, Storing and Processing Light."
 

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
Originally posted by: ericlp
http://money.cnn.com/news/news...marketwire/0375342.htm

This is really good news! This may be the finial chip evolution design the world has been waiting for...

Not trying to be a downer or a thread-crapper here...but my impression is that the world is waiting for are applications (software, not hardware) that will do stuff we don't even realize we want to be doing...like what the internet did for computer adoption rates in the 90's.

There aren't any killer apps for today's MCM'ed quad cores, let alone ones with 100's of cores and on-die optical interconnect.

I would love to have such chips, I program and I use all my quads at 100% utlilization. But I'm not fooling myself into thinking my parents need one, nor my kids, nor my brothers or sister or their kids. I'm glad they buy them though, they help drive up the volume which ultimately drives down the per unit cost when I purchase them too. But they are all waiting for something to come along that compells them to use those cores, let alone something that will use 96 more of them.

Just thinking out loud here, not trying to be a downer. I love researchers pushing the envelope (that was my job too afterall for years at Texas Instruments)...but making a connection between HPC stuff that is itself 5-10 years away from getting put into the worlds most expensive supercomputer versus the 20 years still even further out that it might cost cheap enough to put into a personal computer...and well my excitement starts to fade a tad.
 

ericlp

Diamond Member
Dec 24, 2000
6,139
236
106
Yes, software is important... You see, tho, with an advance like this... This technology I believe will move us out of the (rat brain) to supreme brain (smarter then us)... You have to have a lot of power to do things like that. I believe that this is the tip of the ice burg... Since there is so much data that can be passed in a light wave, This is just cutting edge stuff ... This is like an Intel 4004 stage... Haven't even begun to tap the resources of 8080 / 286 / 386 / 486... When we get to that stage next 10 or so years. I think that the power of this new revolution will have computers designing themselves and coding (self rewriting) err learning software...

I don't see this taking more then a few years before we start seeing real apps... Like I am sure anyone will be buying these for cell phones/PDS and laptops. We should be seeing low powered chip designs that are simple and low cost in cell phones since the power demands of this technology is really good. Not only can it be smaller and run 100 times faster ... but it will draw next to nothing power wise... Imagine a laptop that has no fans... Just think if you laptop didn't produce heat from the CPU/GPU... I think you could get double + power life out of your batteries with this technology. A lot of money is being poured into the Battery technology. This will solve a lot of problems... I think once it stats it will be rather quick to get in machines everywhere as the benefits are amazing....