Photonic Futures

Opinion: The future belongs to multicore processors with multiwave-length optical connections.

My irony alarm started clanging when I read the late-June headline, "Solar sail official blames rockets." Isnt that always the way? You have a really cool new idea, but when you try it out, the parts of the project that you supposedly already knew how to do are the ones that make you (in this case, probably literally) crash and burn. IT people recognize that theme.

For those of you who didnt catch the story about Cosmos 1, a solar sail—in principle—propels a payload through space using the small but quite real and no-fuel-needed thrust obtained by bouncing light off a reflective surface. The idea almost dates back to the 17th century. I say "almost" because Kepler didnt actually know what he was seeing when the tails of comets seemed to be getting blown away from the sun, but he did still make the first known proposal to sail through space.

Thats another irony familiar to IT architects: You can totally misunderstand whats going on but still come up with what turns out to be a viable concept.

Low-tech weakness in high-tech systems, as it happens, was the specific problem on my mind when news of the Cosmos 1 failure sailed across my screen. I was thinking about basic problems of component connection—and about the things that IT inventors do with light—after my conversation near the end of June with Alex Dickinson, CEO of Luxtera.

Luxtera is a fabless semiconductor company thats about one year away from sampling CMOS-process semiconductor chips—initially from the PowerPC builders at Freescale Semiconductor—with integrated optical modulators running at 10G bps and with low-cost optical connection technology.

Luxtera is concentrating on lowering the cost of optical-network construction—unlike Intel, whose integrated photonics effort focuses on chip-to-chip communication—but both applications should benefit from the differences between photons and electrons. "Were dealing with an order of magnitude less power and two orders of magnitude less latency than an all-electronic circuit," Luxteras Dickinson estimated during our conversation.

/zimages/1/28571.gifClick here to read more about Intels photonics effort.

In one key respect, optical interconnection has a big advantage over space flight: Its not very useful to send 10 rockets one-tenth of the way to orbit, but multiple optical data streams using different colors of light can readily be sent along a single fiber to achieve high data throughput without trying to modulate any one signal at hundreds or thousands of gigabits per second. Multicore processors with multiwavelength optical connections look to me like a nice combination for tomorrows high-performance systems.

Thinking further on solar sails, their biggest problem, as no IT person will be surprised to learn, is not in the basic hardware but in its management. You have to unfold the thing, aim it to yield its thrust in the direction you want to go and steer it to maintain that course—without adding massive infrastructure that winds up costing you more than you originally hoped to save.

A project like this doesnt need rocket scientists—you know, those people who think its OK to blow up something just because its not going in the direction theyd planned. A project like this needs IT managers who understand that being off course is normal but that you have to get somewhere useful anyway.

You also have to get your solar sail up and into space, where the force of mere light is actually enough to move things without atmospheric drag to get in the way. Theres another IT parallel: Before we can do the fast, cheap, inhumanly accurate work made possible by digital electronics, we have to get out of the gravity well of data acquisition and away from the constant drag of human error.

If your rockets dont do their part of the job in getting you off the planet, nothing else much matters. Thats what happened to Cosmos 1. As of this writing, the best guess is that the booster rocket stages didnt separate.

In the IT arena, if your data collection devices, network interconnects and even printers dont work correctly, the subtler and more strategic achievements of data mining and Web services orchestration may likewise be moot.

Technology Editor Peter Coffee can be reached at

To read more Peter Coffee, subscribe to eWEEK magazine.

/zimages/1/28571.gifCheck out eWEEK.coms for the latest news in desktop and notebook computing.