At first glance, it’s an unremarkable gadget. About the size of a nickel, the device is made using standard technology borrowed from inkjet printers that squirts tiny bubbles at the intersection of channels carved in a slice of glass. But this seemingly mundane piece of optical equipment performs one of today’s most sought-after technology tricks. As light from an optical fiber shines onto it, the light is guided down one of the channels and, at the intersection, a bubble deflects the light beam, deftly rerouting it to just the right outgoing fiber. This “optical switch” is orders of magnitude smaller than anything now on the market and vastly outperforms existing devices, orchestrating 32 beams of light in less than one-hundredth of a second.
Not impressed? Consider that when Agilent Technologies, a recent spinoff of Hewlett-Packard, unveiled a prototype of the device at a technical meeting in Baltimore earlier this year, the company’s stock soared 47 percent, adding $23 billion to its market value. Such investor exuberance is not limited to Agilent’s version of the technology. Optical switches, which route information in the form of light, rather than converting it to electrons as most current switches do, have become one of the hottest items for those planning tomorrow’s communications systems (See companion story: “Dialing for Dollars”). A week after Agilent’s announcement, Nortel Networks spent $3.25 billion to buy a Silicon Valley startup called Xros that has a promising-but commercially untested-switching technology. And leading manufacturers of communications equipment, including Lucent Technologies and Corning, are making the development of their own optical switches a top R&D priority.
So far, communications systems have managed to keep up because the volume of phone calls, Web pages and videostreams that optical fibers can carry is doubling every nine months, thanks in large part to the ability to squeeze more wavelengths of light into each fiber (See ” Wavelength Division Multiplexing “, TR March/April 1999). It is this remarkable growth in capacity that prevents your favorite MP3 recordings and webcasts from causing gridlock on the Internet.
The problem is that while optical fibers are carrying more and more information, when this speeding light reaches the networks’ central hubs, it must still be converted into electrons and switched by bulky and expensive electronic switches and then converted back to light signals. Converting pulses of information from light into electrons worked fine when fiber optics carried only one signal over limited distances. But electronics have difficulty keeping up as the optical signals become more complex. In the current data maelstrom, the capacity of electronic switches is quickly being outpaced by that of the fiber-optic cables connecting the hubs. In the parlance of optical networking, this data jam waiting to happen is called the electronic bottleneck. “It’s almost a law of nature that any kind of electronic box at the end of the fiber won’t be able to deal with the kind of bandwidth coming out of it,” says David Bishop, director of micromechanical research at Lucent Technologies’ Bell Laboratories.
Enter the new breed of photonic switches-of which at least a dozen are on the way to market. These switches skip the step of converting light into electrons to switch the beams, and, unlike electronics, all-optical devices can deftly redirect even the most complex light streams. All the approaches being tried rely on shrinking the switch components. But after that, it’s a technology free-for-all. Some, like Agilent, employ columns of glass called waveguides to shuttle the light to and from a switch. In the Agilent switch, bubbles deflect the light between crisscrossing waveguides; optical-fiber giant Corning is attempting to use liquid crystals-the same light-bending materials found in your cell phone and calculator readout-to redirect the beams. At Bell Labs, Bishop and his coworkers use arrays of tiny tilting micromirrors to manipulate hundreds of beams at the same time; some 256 of these micromirrors fit on a few square centimeters of silicon.
For those hoping to supply tomorrow’s optical communications equipment, offering some version of these switches is critical to staying competitive. “You either have a technology in this space or you plan to be marginalized,” says Bishop. “There’s a huge amount of money at stake.”
Indeed, network operators are already beginning to install the first generation of these photonic switches, using them much like the mechanical switches in a busy railroad terminal. One immediate advantage is that telecom providers will be able to reconfigure their networks on the fly rather than having to send out technicians to patch a maze of optical fibers. If a backhoe in Des Moines takes out MCI’s Denver-to-Chicago line, tiny tilting mirrors may be all it takes to divert billions of data packets around the break. Operators will also be able to offer dedicated circuits linking a client’s corporate headquarters to its manufacturing plants and customers. In the jargon of telecommunications it’s called provisioning. “Today, it is a manually intensive process that takes months. The ability to do that at the software level is incredibly attractive,” says Yankee Group optical networking analyst Alex Benik.
But that may only be the start. As photonic switches become even smaller, more functional and cheaper, many experts believe they could help bring the massive bandwidth of the network core closer to end users. “As this technology becomes more readily available, the high-speed pipes will extend further and further into the network,” says David Andersen, R&D director for Agilent’s optical networking division. “As that bandwidth becomes available the experience of the Internet will become richer and richer.”
Experts are quick to point out that even today’s most impressive optical switches represent only an early step in the evolution of functional optical devices. In a world of advanced microelectronics, today’s prototypes are still bulky-and they’re dumb. Although they can switch light beams with ease, they can’t read the messages carried in them. As a result, electronic devices must be used to control the photonic switches, telling them when and where to redirect the beams of light. That’s acceptable if the goal is simply to reconfigure the main pathways of light at the network hubs. But that initial goal certainly will not satisfy the ambitions of telecom providers for long. They would like to extend the full power of optical switching all the way to the end users, creating a truly “all-optical Internet.” And that means smarter optical switching.
The Internet is built around packets-strings of data that find their destination by hopping between local and regional “nodes” where the mesh of transmission lines that form the Internet intersect. E-mail a snapshot and it will be chopped into hundreds or thousands of packets that travel separately from node to node. Each packet carries with it an Internet protocol, or IP, address that is read by electronic switches called routers at each node. Whereas the switches at the network backbone can blindly shift an immense number of packets in bulk, routers must ponder each packet. The IP address tells the routers where the packet is going; the routers then forward it to the next appropriate node. This process is repeated across the Net until the individual packets that make up the snapshot have arrived at their destination, where they are reassembled.
This forwarding arrangement is part of what makes the Internet resilient. If a node goes down or is temporarily swamped, neighboring routers simply divert the flow of packets around it. However, routers are straining to keep pace with the surging Internet traffic, just like their electronic cousins in the network backbone. Each channel of light in today’s fastest fiber-optic cables transmits thousands of packets every second, with additional channels being squeezed in all the time. While electronic routers have managed to keep up with converting the optical signal to electrons and reading the IP address on each packet, they are now being pushed to their limits.
Herein lies the next big job for photonic switches. In tomorrow’s all-optical Internet, operating at speeds unimaginable with electronics and delivering that speed to your home PC, photonic switches must be smart enough to serve as routers by recognizing the packets of information and determining where they should go. In other words, the devices must have the speed of optics and the intelligence of microelectronics. And that will happen, say those in the field. “The future of networking is going to be optical IP switching,” says Gary Bjorklund, chief technology officer with Miami-based Nanovation Technologies.
To spur the creation of optical components that are far smaller and smarter, earlier this year Nanovation pledged $90 million over six years to finance research at MIT in the field of microphotonics. The goal is to emulate the integration of transistors and other electronic devices onto a chip by shrinking optical switches, fibers, lasers and detectors, and stringing them together on a single optical circuit. Just as stuffing together more and more transistors, diodes and capacitors has greatly multiplied the speed and power of microelectronics, both MIT and Nanovation are betting that integrated microphotonics circuits will deliver similar advances in performance.
Nanovation’s short-term interest is in delivering smaller, more efficient network devices, including simple switches (this spring the company introduced an initial line of such products), but down the road the company hopes integration will deliver more sophisticated and functional optical devices. “This is like the very early days of integrated circuits,” Bjorklund says. “The day after they invented the integrated circuit they didn’t go from individual transistors to the Pentium. That took thirty years. But there is a lot of value added by putting twenty or thirty transistors on one chip.”
Before Nanovation-or anyone else-can become the Intel of microphotonics, they must learn to tame the unruly photon. Electrons have proven useful because they have a charge and are therefore relatively easy to manipulate by, say, applying a voltage. That makes it possible to route electronic signals around microscopic integrated circuits; it also makes it possible to utilize electrons to manipulate and store information. It is these qualities that have made electrons so valuable in microelectronics and computing.
Photons, however, are much less malleable characters than electrons. They’re great for speeding information from one place to another over long distances, but it’s tough to control them. Try sending them around a sharp corner, for instance, and they scatter wildly.
Dreams of taming photons and using them in integrated optical circuits are not new. But recent research in industrial telecommunications labs, universities and at least a dozen startup companies is opening up ways to better invoke the required mastery over photons.
One key to this work could be the fabrication of smaller and more efficient waveguides that could be used to control the flow of photons in integrated optical circuits. To gain higher levels of functionality for optical circuits-and gain a higher level of control over photons-Nanovation and MIT, among others, are exploring waveguides fabricated in novel materials such as silicon-silica hybrids and indium phosphide. These materials confine light so strongly that beams with a wavelength of 1.5 micrometers-the standard for telecommunications-can be contained in waveguides about half a micrometer wide. These waveguides can also make tighter turns-about a thousand times tighter than the millimeter-scale bends achieved in earlier waveguides-without spilling their contents.
Photons confined in these tiny structures take on odd and potentially useful behaviors. For example, relatively small voltages can entice photons in the waveguide to jump between adjacent channels. And rings as small as 5 micrometers in diameter-about the width of a human hair-can pluck lightwaves of a given wavelength out of one waveguide and shoot them down another, providing a color-selective switch.
Such advances are opening up possibilities for integrated photonic chips. And that is boosting hopes for smarter optical switches that will better distribute the power of fiber optics. For instance, an optical chip could select one wavelength of high-bandwidth light off the fiber passing through your neighborhood and divert that data-rich wavelength to your home, where another optical chip would grab your phone calls, high-definition TV shows and Web browsing. This type of networking is already being explored in regional systems. Future microphotonics could bring it to the living room.
The promise inherent in integrated photonic circuits, however, extends beyond fiber-optic networks. In much the same way integrated optical devices could improve how data is transmitted in long-range communication systems, they could also facilitate the sharing of information in the microscopic realm where computers talk to themselves. For years physicists have been dreaming of computing that relied on photons, rather than electrons. That idea may or may not become a practical reality; while it has proven possible to use photonics for logic and memory functions, electronics remain much better at the job.
Still, photons may be the key to faster computer chips, which are suffering from their own form of electronic bottleneck. Increasingly, the speed of transistors and integrated chips are outpacing the metal wires and cables that connect them-the wires can’t carry data fast enough. “Whereas microphotonics is giving you complex optical circuits for communications, now it’s also required at the other end-the very localized end-to make the next generations of microprocessors,” says MIT’s Kimerling.
Wires of Light
Photonic wiring between transistors could make faster chips, while photons zipping between chips could turbo-charge your computer’s motherboard. Imagine thousands of laser beams crisscrossing within your computer, and you wouldn’t be far off.
Engineers at companies such as Honeywell, Sun Microsystems and IBM, as well as at universities around the world, are already testing arrays of light-emitting diodes and laser beams that could serve as the “bus” that transports information across a motherboard from microprocessors to memory chips to display screen and back. Initial applications for these systems will be in high-end computing. Honeywell, for example, is using optical interconnects to link microprocessors, creating compact, powerful parallel computers.
Squeezing light inside individual computer chips to replace metal interconnects may not be far behind. The ability to etch smaller transistors on silicon wafers has steadily increased the power of computers. The problem is that as transistors get smaller and smaller, they can switch much faster than they can communicate with each other, slowing down the overall functioning of the chip. While chip makers think they can squeeze out enough performance from the shrinking metal wires that connect electronic devices to make the next generation of chips, they are turning to photons for future batches. Sematech, the international association of semiconductor makers, estimates that its members will begin to exhaust improvements to metal interconnects by 2008. The use of light-based interconnects is one of the few feasible solutions, according to Sematech.
Optical interconnects in computing “is still speculative,” says David A.B. Miller, an electrical engineer at Stanford University. But, he predicts, you’ll see such interconnects “in mainstream computing by the end of the decade. That’s not a ridiculous statement to make.”
Whether working on telecom or computing applications, however, the ultimate goal of optics researchers is to make integrated microphotonic circuits as ubiquitous as today’s microelectronic chips. “The ideal solution is to have something able to do the mirror function, the switching function and the waveguiding function all within one platform,” says Shawn Lin, a physicist at Sandia National Laboratories in Albuquerque, N.M. Lin is using photonic crystals, tiny silicon-based structures that confine light with exquisite efficiency, to make his own set of lasers, amplifiers, waveguides and other devices for lightwave circuits. Getting all of these devices to work together on a single wafer, says Lin, will put photonics on “holy ground,” just as moving the electron from a vacuum to the silicon wafer unleashed the power of electronics. What’s more, Lin thinks integration will take hold in photonics much faster than in microelectronics.
Lin’s voice falls to a whisper as he discusses the future of microphotonics, as if revealing the full potential of capturing light on a chip could somehow jinx his chances for success. “All the necessary inventions, the materials issues have been solved,” says Lin. “It’s up to our imagination to come up with innovative devices and to build the basic building blocks. There are large amounts of money just waiting to see breakthroughs happen.”
Those breakthroughs are likely to come from the dozens of industrial and university labs that are working on integrated photonics, as materials scientists like Lin work with optical theorists to perfect waveguides, lasers and other basic building blocks, fabrication experts figure out how to integrate these devices in a high-density chip, and systems engineers optimize circuitry. Just as today’s micromirrors and tiny bubbles are beginning to switch on the full potential of the Internet’s high-speed optical backbone, tomorrow’s optical chips promise to unleash the photon’s raw power and speed to fundamentally change how information is transmitted.
Building Tomorrow’s Optical Internet Company Recent Moves Strategy Alcatel No major acquisitions Already a leading player in DWDM, the telecom giant is spending heavily on R&D to develop optical devices Cisco Systems Bought ArrowPoint Communications (May 2000) for $5.7 billion; Pirelli Optical Systems (Dec. 1999) for $2.15 billion; Monterey Networks (Aug. 1999) for $500 million Gained Internet switching technology from ArrowPoint; DWDM technology and optical switching devices from the Pirelli and Monterey acquisitions Ciena
Acquired Lightera Networks (March 1999) for $552 million An early pioneer in DWDM, looking for a business comeback by providing intelligent optical switching Corvis Testing its all-optical devices with Broadwing Communications Leverage technology for ultra-long optical transmission and optical switching to provide an all-optical network Lucent Technologies Micromirror-based optical switch is scheduled to be installed this year Relying on internal R&D to develop optical devices; a leader in microelectromechanical systems Nortel Networks Bought Qtera (Dec. 1999) for $3.25 billion; Xros (March 2000) for $3.25 billion; CoreTek (March 2000) for $1.43 billion Spending heavily to acquire the necessary technologies to put together an all- optical Internet
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
Data analytics reveal real business value
Sophisticated analytics tools mine insights from data, optimizing operational processes across the enterprise.
Driving companywide efficiencies with AI
Advanced AI and ML capabilities revolutionize how administrative and operations tasks are done.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.