By Burt Helm Telecom geeks tend to revere old gear -- from copper phone wires to mechanical switches -- as vintage collectibles. Punch in "Western Electric" -- the name of AT&T's (T) former manufacturing unit -- on eBay (EBAY), and you'll get close to 500 offerings of relics of the communications systems of yore.
Such antiques stand in sharp contrast to today's networks, which turn voice, data, and video into light and then speed it across the thinnest of glass fiber-optic cables. Such super-efficient systems depend on complicated and expensive electronics to convert the signals from electricity to light. But new breakthroughs in silicon-based optoelectronics -- also known as photonics -- could make fiber systems cheaper and even more efficient and accessible in just a few years.
The most recent breakthrough, announced last month by Cornell University researchers in the scientific journal Nature, is a switch that converts light into an understandable electrical signal more quickly than ever. Today's optical networks are actually hybrids. Any information -- be it a phone call or an e-mail -- zooms down optical wire in the form of photons. But when it gets to an intersection, the light has to be converted into an electrical signal and programmed with instructions on where to go next -- which slow the processes down.
SILICON DREAMS. Researchers at Cornell say their all-optical switch would be able to direct signals without converting them into an electrical signal, eliminating one step in the process. This could radically enhance the speed and cost-efficiency of the telecom network, they say.
And more developments in optical communications are on the way. Scientists at Intel (INTC) and IBM (IBM) are exploring how to use silicon wafers -- so successful in bringing about the rise of smaller, cheaper, and more powerful computers -- to communications electronics.
Researchers hope that one day they'll be able to use cheap, easy-to-manipulate silicon wafers to program the lasers that are the heart of today's optical networks. Those lights transmit information by flashing on and off in Morse code-like patterns, which are controlled by large, complex machines.
FASTER DATA THAN EVER. Using chips etched into silicon wafers to turn the lasers on and off would cut production costs of the signal transmitters and receivers. "Ten years ago, very few people were looking at silicon photonics," says Vic Krutul, manager of the silicon photonics division at Intel. "It has really only been recently that people have said we can see [such] a device in the future."
If optical electronics could be made cheaply enough, computer-hardware experts believe they could then have powerful applications in supercomputing, which could enable huge data transfers with little power output. So far, fiber optics have only made economic sense acting as a major artery in supercomputers, while copper wiring makes up the rest of the connections.
But what would happen if all the connections in supercomputers -- which are really networks of very powerful machines -- were made with fiber optics? Data could move faster and more freely than ever. Faster information transfer could improve weather forecasting, designing and evaluating drug compounds, or creating more aerodynamic cars, for example.
"GETTING THE RIGHT RECIPE." The information needed to crunch all these calculations would be shared across networks that use optical switches, just like the ones under development at Cornell and also in Lucent's (LU) Bell Labs, which is working on a competing, non-silicon switch design.
True, these developments will take time to come to market. "All of [the current problems] are fairly well understood, but it's a matter of getting the right recipe that will optimize the components," says Krutul of Intel's optical electronics. Once the recipe is perfected, the ability to "rubber-stamp" the designs inexpensively will come very quickly, because they are being designed specifically to use current semiconductor-manufacturing processes.
Challenges remain, however. Researcher Modest Oprysko at IBM stresses that although excellent progress is being made with optics, reliability issues will need to be solved. Most importantly, optics still will need to compete with advancements in other areas, particularly in copper.
TIME LINE: 6-10 YEARS. "We can go out and buy a gigabit Ethernet connection [which uses a copper/optical hybrid for transmission] right now for about 50 bucks -- it's hard to compete with that," says Oprysko.
Intel predicts it will take about six years to develop silicon wafers to manage optical networks. Cornell's all-optical switches won't be ready for market for at least another decade. But that's a blink of the eye, historically speaking. Once these new devices are ready, hang onto all those amplifiers and lasers in the back room of your company's information-technology department. They might be worth something on eBay one day. Helm reports for BusinessWeek Online in New York.