IBM's Holey Optochip capable of 1 terabit per second transfer speeds

Shawn Knight

Posts: 15,240   +192
Staff member

Researchers at IBM have demonstrated a prototype optical chip that can transmit up to 1 terabit of data per second. Engineers have managed to build the Holey Optochip using components that are readily available today, meaning the chip could be manufactured in high volumes for commercial implementation sooner rather than later.

The Holey Optochip was envisioned as part of IBM’s continued efforts to use light rather than electrons over wires to transmit massive amounts of data. The chip was created by taking a standard 90nm CMOS chip and drilling 48 tiny holes through the back of it. This allows access for 24 receiver and 24 transmitter channels, each capable of moving 20 gigabits of data per second.

The end product can move 960 gigabits (nearly 1 terabit) of data per second, speeds that could transfer 30,000 HD movies each minute. A single chip could transfer the entire Library of Congress in just one hour - all of this from a chip that is one-eighth the size of a dime and is powered by less than 5 watts of electricity.

Revisions are already in the works that increase channel output to 25 gigabits per second which would produce a chip capable of 1.2 terabits per second. IBM Optical Links Group manager Clint Schow told Ars Technica that IBM won’t be mass-producing the chips but they could become commercially available within a year or two for between $100 and $200 each.

IBM will be presenting the Holey Optochip at the Optical Fiber Communications Conference in Los Angeles today.

Photos courtesy Ars Technica.

Permalink to story.

 
Is this is CPU?
im confused. Assuming it is,
I CAN HAS?
but won't it be throttled by hard drive and stuffs?
 
That is very low power consumptions, dang it whoah 5W? Id do my homework but its kinda late, so how much more powerful would it be compared to lets say i7 2600k or lets say X5690?
 
but won't it be throttled by hard drive and stuffs?

Was thinking the same thing as well...Still the most important tech im looking forward to this decade is Quantum Computing and Nanotech, those are the two HUGE technologies, will change the World big time.
 
ikesmasher said:
Is this is CPU?
im confused. Assuming it is,
I CAN HAS?
but won't it be throttled by hard drive and stuffs?

This is a interconnect chip, not a general purpose CPU, or even a network interface. Although I don't see why the technology couldn't be applied to networking. No distance limitations are mentioned however.

When you start dealing with computing clusters, you can't fit everything into one chassis, so you need super-fast ways to transport data between individual systems. Think of how fast your CPU can talk to your RAM, in a Sandy Bridge-based system, about 37GB/s. This interconnect is capable of 128GB/s ... as long as the latency is low, you're remove the chassis to chassis bottleneck.
 
I never knew that by drilling holes in a chip will make it faster. I shall try that on my desktop processor.
 
"The chip was created by taking a standard 90nm CMOS chip and drilling 48 tiny holes through the back of it. This allows access for 24 receiver and 24 transmitter channels, each capable of moving 20 gigabits of data per second.

The end product can move 960 gigabits (nearly 1 terabit) of data per second, speeds that could transfer 30,000 HD movies each minute."

Ok, so... is what they did like cutting a shortcut through a hedge maze if you didn't want to bother going through all the BS twist, turns, and dead ends? Because if so... why did nobody think of this sooner? o_O
 
Guest said:
I never knew that by drilling holes in a chip will make it faster. I shall try that on my desktop processor.
DUDE, tell us how it goes, Its gonna work SO well.
 
How about a CPU on Optochip design? Or rather a motherboard circuit with fiberoptic engraded on (rather in) it??? hmmm...
 
LinkedKube said:
Oh come on, they know they wanted to name it "HOLY" chip.
Ah, yes, but they had Brother Maynard consult the Book of Gadgets, which clearly states:

"Thou shalt not take the name of thy Holy Optochip in vain, for it bestows many computations to thee in its Holy logic." - Gadgets 1:5
 
Now just figure out how to Interface this with the human brain and people could go from zero to stupid at 1.2 terabits per second.
seriously, can't wait to see how and where they'll Implement this first, I'm all tingly with excitement.
 
Wow... First of all this chip is not a typical chip its does calculations with light also known as optical computing. This and several other advancements namely this one (http://www.mit.edu/newsoffice/2011/optical-computing-diode-1123.html) from MiT. It is still a new technology and more or less a idea. But it is moving along really quick and its very exciting.
 
This is good as in interim solution in datacenter and that too IBM. I do not see other company will follow suit. And I do not see it being adopted in user systems. The only commercially available soln in the near future will be intel Light peak and they too is stepped down to use copper link since fiber optic solution is not commercially viable solution for service providers unless they upgrade their infrastructure.
Or
The ISP can use these kind of links between their various nodes and offer W-MAX kind of connectivity to reach end user. They me work better for user to get the benefit of new improvements. Since any new chip that comes in needs to be accepted and adopted by the ISP and telecom companies first so the user can then get benefit of it. So no point in major research or break through unless you have some major players like AT&T, Sprint, (US); BT (UK); China Mobile (China); Airtel, YOU, Hathway etc (India) getting on board with the implementation.
First sign up the ISP then do the recherché and then they will adopt the new tech.
 
If this chip needs 48 individual fibers to achieve the bi-directional terabit speeds it claims (not multiplexing the transmit or receive channels to a single fiber), then this is not new technology. For the price, it could probably be a good solution but needing 48 fibers for a terabit bi-directional interconnect is expensive. If anyone is thinking that this will compete with the latest 40 and 100 gigabit transport standards, think again. This isn't even probably carrier grade. The industry's best optical chip is capable of doing 500 gigabits on just two fibers - 48 fibers would yield 12 terabits - plus its designed for carrier-class networks and applications.
 
Back