Exabit Transmission Speeds May Be Possible

adeelarshad82 writes "Scientists at UC Berkeley were able to shrink a graphene optical modulator down to 25 square microns in size (small enough to include in silicon circuitry) and were able to modulate it at a speed of 1GHz. The researchers say that modulation speeds of up to 500GHz are theoretically possible. According to the research, due to the high modulation speeds, a graphene modulator can transmit a huge amount of data using spectral bandwidth that conventional modulators can only dream of. Professor Xiang Zhang, in an attempt to boil his group's new findings into consumer-speak, puts it this way: 'If graphene modulators can actually operate at 500GHz, we could soon see networks that are capable of petabit or exabit transmission speeds, rather than megabits and gigabits.'"

Comments

Faster than silicon

[empiricistrob]

So in theory if you can get an electrical signal to the graphene, you can use it to modulate laser light up to 500ghz. Awesome!

That just leaves two fatal flaws:
1. You need to modulate the electric signal with useful information at 500ghz. I'm not an expert, but it seems like we're a long way off from being able to do that. Can anyone comment?
2. How do you demodulate such a signal?

Re:Faster than silicon

[drolli]

1. there is a logic which is nearly fast enough. It's called RSFQ, but interfacing it to graphene may be difficult.

2. with RSFQ ADCs.

If its about analog mixing, you could use bolometer mixers, interfacing to RSFQ circuits.

Re:Faster than silicon

[mpoulton]

The modulation problem can probably be solved with clever use of current technology. Initially at least, the only application for links with this bandwidth would be in aggregated data transmission, accumulating dozens or hundreds of lower bandwidth connections. A clever modulation method would utilize multiple separate electrical modulation signals to control the optical modulation, possibly by using multiple separate modulator elements in the optical path, each operating at a lower modulation rate (but synchronized with the others and phase shifted). In the long run it will be interesting to see how data transmission technology evolves to accommodate high data rates like this. 500GHz is hardly even an electrical signal, it's almost light-like. Wires don't work at those frequencies; it's waveguide-only territory. It can really only be handled easily as a modulated optical signal. If we are to progress to a point where data rates like these are practical for individual computing devices we will have to switch to all-optical protocols for networking, and probably also for internal data transport within computing devices. Demodulation of an optical signal with this much modulation bandwidth is pretty much an unsolved problem for now, AFAIK. As with the modulation process, I'd probably try to split it into multiple channels each covering a narrower bandwidth. Unlike the modulation process, I can't think of an obvious way to do that off the top of my head. It's also worth noting that the professor seems to be contemplating the use of many optical modulators (each at 500GHz), each operating on a different fundamental wavelength to multiply the link bandwidth. Hence the prospect of petabit and exabit data rates from 500GHz modulation.

Re:data storage?

[Puff_Of_Hot_Air]

It's all well and good having super fast transmission capabilities but do we have anything that can process/store data as quickly? It's an honest question as I've always been lead to believe that data storage is the bottleneck.

Infrastructure is where this is important. There are these extremely expensive cables made of glass under the ocean connecting various land masses. It's extremely convenient to be able to upgrade the boxes at either end instead of laying more tubes (*warning* simplification!). You don't need to store the data (at least not in one box), you just need to switch it. This is why fiber is so awesome; people just keep on discovering new ways to jam more down those pipes!

Not at all levels

[Sycraft-fu]

You have to remember that the more bandwidth you want to deliver to the end user, the more you've got to have in the backhaul. Like if at work you want to deliver true 1 gigabit to 1000 people's desktops, you can't very well then have a 1 gigabit connection out to your data center. They won't get a gigabit of performance.

So while speeds like this wouldn't be needed for servers or such, they could be for big links. You want to link big_router_a with big_router_b which have all sorts of very fast connections to smaller routers then maybe this interests you.

Re:data storage?

[erayd]

Actually, if the Internet backbone ever reached exabit/sec speeds, the whole way we viewed data storage would change. I see no reason why most computers would need local storage at all.

Latency is why. It doesn't matter how fast the link to your storage is, if it's several ms away from you the delay gets annoying *real* fast.

Re:Pointless?

[AlecC]

Many systems connected to many systems. It doesn't have to be to/from a single CPU or storage device. You can put your datacentre where it is most efficient in energy or cooling terms, but have it appear to be where you want it operationally. Or you can aggregate datacentres scattered across the globe into a unified system, load sharing as the peak load moves round the globe. It makes the physical attributes of "the cloud" more possible.

Re:Pointless?

[vlm]

And specially storage speeds. SSDs don't cut it.

Oh of course they do. You just have to use more than one, in parallel / striping mode. Think of a "real" NAS or an IBM DASD with dozens of drives in parallel.

Probably this will be used mostly for DWDM style stunts... Find the fastest system and its press release. Insert two in a box twice as big. Issue press release to the mass media, and sadly, /., reporting "new world record of twice the libraries of congress per second". The general public responds with "who cares" because that kind of press release is issued seemingly daily or weekly because its so easy to put more copies in a box or a rack or a couple racks. The techies respond with "who cares" because solving the power and thermal problems of "put two in a box instead of one" isn't very impressive, and its about as dumb as putting two AM radios in a box and calling it "AM stereo".

Dispersion! But On-Chip Networks...

[Talisein]

I haven't done the math, but at 500 GHz it seems like dispersion would make any network longer than a single chip fundamentally unable to use that kind of frequency.

For a mesh network-on-a-chip though, you could probably dumb down the routers a lot (you'd have to to let them operate at that freq), and basically trade inefficient routing for a way higher link rate... basically operate the network such that you can deliver a message 100 times faster than than you can send 1 message. The routers may not even need buffers at that point. But I think there are a lot of problems here.

I think the parent comment is right: 500 GHz modulator is nice and all, but its difficult to use until everything else is at least on the same order of magnitude.