A Working Light-Based Processor

Breakthrough light-based microprocessor chip could lead to more powerful computers, network infrastructure

www.colorado.edu...< br />
www.digitalnewsarena.com...

www.nature.com...

Snippet from Digital News Arena:

A new type of processor has been developed by the researchers from the University of California, Berkeley, one that uses light in order to transmit data. Together with other universities, UC Berkeley experts have created a revolutionary, world’s first processor that fully works with light. This cutting-edge processor is much more advanced than conventional electric chips, having the ability to process 300 Gbps per square millimeter. Compared to a standard processor, that’s 10 to 50 times more data. The new chip measures 3 by 6 millimeters, and it features 2 processor cores with 70 million transistors. In addition to that, we have 850 photonic components, which are used to receive and send light.

I frankly am stoked. I am hoping these become commodity CPUs in the very near future. Can you imagine when this is paired with photonic memory chips: sciencebeta.com...

Machine Intelligence will soon be leaping forward by a factor of 50X.

Sounds nice, the production of them seems simple as a lot of it is using the same tech as now the only problem i see is probably about 20 years of patent laws as everyone tries to get a cut of the pie. Lower thermal will be what will drive interest as in data centres its amazing how much cooling costs for high density servers so just the fact you can drop down by lets say 30% your cooling bill may be enough to grab interest.

a reply to: machineintelligence
It's exciting, but like quantum computers and other magical tech, it's starting to make my eyes glaze over when I read about it. I can't comprehend what it really means. And by that I mean I haven't read all of the material and I don't know exactly how practical it's at the moment--so what it means to somebody who actually works in the industry at present.

Some other links regarding light-based technologies:
qz.com - Scientists can now “squeeze” light, a breakthrough that could make computers millions of times faster...
phys.org - New technology enables ultra-fast steering and shaping of light beams...
www.spacedaily.com - Electrical engineers take major step toward photonic circuits...

I believe you're right when you say AI will get 50x speed up over the following years. But I think it'll be much more than that, also due in part to improvements in the code. A lot is changing for sure.

I think we'll be commonly using our devices and they'll be acting semi-intelligently. The only question is, with machiens being able to do so many physical and mental labors, what will that leave for us? In the past people wondered what people would do if machines could do all of hte heavy lifting? A lot of old folks blamed machines for killing jobs. Another thing often left unsaid is the societal changes. For example, men were commonly filling those physical jobs and now that machines are increasingly doing them, men are having to find other work. (I wonder sometimes about that. Maybe men are impacted more than woman by this.)

I was reading about history yesterday, trying to make a post about infromation technologies. And it occured to me the world has always been undergoing these changes. The past IS a guide to what to expect, except Kurzweil might be correct in some way about the exponential pace of technological change. I just think he's not entirely right. Hillary Clinton said civilization moves at a snail's pace and she was right. I also believe there're unanticipated limits which constrain any potential singularity from occuring, the sort Kurzweil likes to argue in favor of.

This seems to be the interface to the external world. For a modern CPU, that's around 1000+ pin connections. Are all of those going to be replaced with optical connections? The network hardware industry would seem to be the first to gain from this. The fibre-optic cables could go straight into the message processing chips. Maybe aircraft systems could benefits from being able to replace all that copper cable with fiber-optic cable.

It also looks like they are using a single substrate that contains the CPU, memory, and IO, all tied together with phototonic interfaces. And, unlike electronic interfaces, they can multiplex a single channel using different wavelengths of light.

But, apparently the biggest breakthrough is that they have figured out how to do this using current manufacturing methods. That will definitely increase the speed at which this technology can be implemented for real-world use.

At one point it looked like we were reaching a limit to the Moore's Law processing increases that we have enjoyed in the past because of the limitations we were reaching in actually creating the masks used to make the chips. New technology like this will not only extend the life of "Moore's Law", but will increase the rate at which these cycles occur.

Cool.

-dex

a reply to: jonnywhite

This optical processor prototype only has the transistor count of a mid range Intel Atom processor or ARM processor but if it can process at 50X the speed of electron based processors it need not scale up much more but just produce more cores and link them directly to photons based storage and optical fiber connects it all together. It will dramatically improve the entire network as well as the individual device or server farm.

originally posted by: stormcell
This seems to be the interface to the external world. For a modern CPU, that's around 1000+ pin connections. Are all of those going to be replaced with optical connections? The network hardware industry would seem to be the first to gain from this. The fibre-optic cables could go straight into the message processing chips. Maybe aircraft systems could benefits from being able to replace all that copper cable with fiber-optic cable.

The problem with probably bringing in fibre directly is the problem of vibration as i've seen fibre optical cables sometimes move just enough when clamped down properly due to local vibrations by traffic etc to lose the odd packet here and there, so while its not a problem for a network to have the odd packet get lost it certainly will be for a CPU.

And as for aircraft from what i remember is the fact that the pilot is a lot slower than the communication medium anyway so theres no reason for control but for extra data to the pilot then maybe.

a reply to: DexterRiley

If you look at the references they site in the paper published in the Journal Nature. Source: ieeexplore.ieee.org... it has some interesting performance data.

Snippet:

Corona is a 3 D many-core architecture that uses nanophotonic communication for both inter-core communication and off-stack communication to memory or I/O devices. Its peak floating-point performance is 10 teraflops. Dense wavelength division multiplexed optically connected memory modules provide 10 terabyte per second memory bandwidth. A photonic crossbar fully interconnects its 256 low-power multithreaded cores at 20 terabyte per second bandwidth.

If this is a commercialization of the Corona architecture it will be pretty awesome.

a reply to: machineintelligence

This is amazing. It's just the start, but light-based technology in layman's terms is transmitting data via it's own energy source. Imagine your email creating it's own path to destination without the help of hard-wired electricity. That's what this is!

a reply to: stormcell

This seems like a commercialization of the Corona Architecture PDF: pages.cs.wisc.edu...

This paper gives a good lesson in the concept of multi-unit photonic compute cores.

originally posted by: machineintelligence
a reply to: stormcell

This seems like a commercialization of the Corona Architecture PDF: pages.cs.wisc.edu...

This paper gives a good lesson in the concept of multi-unit photonic compute cores.

Thanks. That is fascinating. They've brought light communication right onto the chip surface itself so that the multi-core interconnects (cache and network controllers) can communicate in multiple wavelengths of laser light using MEMS technology to detect, modulate and inject light pulses instead of electrons. Underneath all of the traditional electronic cores, is a Hilbert space curved arranged semiconductor crossbar that propagates unicast and broadcast messages to all cores. And all of this is achievable using existing chip fabrication plants, so any type of CPU or GPU could be designed in this way. There isn't a problem with a laser light source because it's possible to implement nano-laser using MEMS:

www.rle.mit.edu...

Imagine doing this with a GPU with thousands of cores... and eliminating tremendous amounts of heat, increasing chip density, and then extending this into superscalar rack mounted supercomputers.

a reply to: stormcell

Interesting stuff going on with quantum cascade THz lasers. With regard to the architecture of the on chip optical information handling, I think how they are using ring resonators for signal I/O and more is fascinating technology. The single photon wave-guides and translating phonons into photons and back again I find that stuff to be bleeding edge. The THz end of the optical spectrum is showing use in generating quantum entangled photons and phonons at data rates that place them above anything currently on the horizon. Quantum computing will be enabled by this architecture approach shown in the Corona 3D optical system.

a reply to: Maxatoria

Huge breakthroughs or findings in science are tend to be held back by one of two things: Military applications coming first and the cool civ-sec tech only coming much later, or corporations wanting their slice of the pie.

It's why I'm not really looking forward to Cold Fusion. Even if it isn't sabotaged by other means, my god those fossil-fuel based energy companies are going to do everything in their power to cockblock the transition.