The Light Brigade

A secretive Silicon Valley startup has achieved a startling breakthrough: a chip that controls the flow of light. Could this be the dawn of the next Intel?

By Om Malik

March 13, 2002

At the end of last summer, three entrepreneurs were crammed into a tiny cubicle. Despite the bone-chilling air conditioning, beads of perspiration dotted their foreheads. Just minutes earlier, they had received a special package and unwrapped it with the haste of children on Christmas morning. Their eyes focused on a gray semiconductor lying inside a clear jewel case. They all pondered the same question: would this chip–one with circuits that control the flow of light, rather than, as other chips do, the flow of electrons–work?

Over the next few days, Jagdeep Singh, Drew Perkins, and David Welch waited anxiously while other engineers in the company thoroughly tested the chip. Four months earlier the three men had founded Infinera and created a design lab in Cupertino, California. And already they had developed their first chip–in about half the time of a typical chip-design cycle. If this chip performed as they hoped, it would shatter many of the theoretical limits regarding the behavior of light in optical communications networks.

When the scorecard finally was tallied, the chip performed better than expected. “High fives all around,” recalls Mr. Singh, the company’s president and CEO. “Dave [Welch] even joked that we hadn’t set the bar high enough.”

Hardly. Infinera’s thumbnail-size chip is the first integrated photonic circuit. Though Infinera won’t reveal the chip’s cost, when built with manufacturing techniques used by chip makers like Intel and Advanced Micro Devices, it likely could be made very cheaply. The savings in manufacturing in turn would lower the cost of network equipment by half, perhaps even more. Beleaguered network carriers like Level 3 Communications and bankrupt Global Crossing could build networks for much less, and run them more efficiently and at a lower cost–maybe even profitably. For consumers, Infinera’s chip could be instrumental in allowing communications companies to offer high-speed Internet access at affordable prices. And one day this technological breakthrough could lead to a device capable of projecting a holographic display, as on the TV series Star Trek.

Such an application is far off. Infinera’s chip first will tackle the vexing problems of optical networking. Today, optical-networking gear–dense wavelength division multiplexers (DWDM), which split light into its component wavelengths for additional bandwidth; regenerators, which provide light waves with a “boost” to travel through hundreds of miles of fiber-optic cable; and add-drop multiplexers, which help ensure no light particles are lost in transmission–are built from a combination of discrete components like lasers, modulators, amplifiers, transmitters, and electronics. The size of these components varies from as small as a pack of cigarettes to as large as a videocassette. Each component performs a specific task, like converting electrical signals into light particles, splicing light waves into different channels, or altering the frequency of light waves. The components are connected to one another by ultrathin optical fibers. When finally cobbled together, these components comprise a device about the size of an air conditioner.

The integrated photonic circuit developed by Infinera, however, shrinks these discrete components down to microscopic levels and combines them on a slice of indium phosphide (InP)–instead of silicon. Infinera says its chip can control light waves, or photons, at micron levels–roughly 1/50 the width of a human hair–and can be modified easily to fit into any piece of optical-networking equipment.

Infinera’s chip will act as the brain of a highly versatile new type of network device that is the size of a pizza box. Just as a microprocessor enables a PC to be used for writing documents, number crunching, or designing greeting cards, Infinera’s chip has the potential to perform many tasks.

Optical Infusion

A typical optical network has five basic functions: combining electronic signals from many sources into photon streams, splicing and rejoining light waves into different channels, transmitting light waves, receiving light waves, and converting photonic signals into electronic signals, each going to a different destination. Other than revealing that their device will perform one or more of these functions, Infinera executives are keeping a tight lid on the features being crammed onto the chip.

It is a lot to expect from a single chip, but the company’s investors are impressed so far. In November, Infinera closed a $50 million second round of funding, bringing its total to $86 million, with investors valuing the company in excess of $250 million. Its corporate and venture capital backers include Applied Materials, Juniper Networks, Accel Partners, Benchmark Capital, Kleiner Perkins Caufield & Byers (which provided a $1 million seed investment), Venrock Associates, and Worldview Technology Partners. Serving on the company’s board are Pradeep Sindhu, founder, vice chairman, and chief technology officer of Juniper Networks; Dan Maydan, president of Applied Materials; T.J. Rodgers, president and CEO of Cypress Semiconductor; Alex Balkanski, a general partner at Benchmark; and Vinod Khosla, a general partner at Kleiner Perkins.

Infinera is in a race with the nearly 700 startups vying for the top spot in the optical-components space. Any one of them secretly may be developing a similar product. For the time being though, Infinera is in the lead.

To avoid losing the race to a company with the so-called second-mover advantage, Infinera has to figure out a way to mass produce its integrated photonic circuits effectively, then develop a device based on its chip, and finally, write software that makes the device work with existing networks.

As if that were not enough, Infinera also faces the challenge of creating new products for a depressed telecommunications market in which companies like Global Crossing and McLeodUSA are in bankruptcy. Carrier spending on new equipment is down sharply and is expected to stay that way for years to come.

Network Connection

So far, however, a combination of luck, ambition, and ingenuity have favored Infinera. Two and a half years ago at the annual Kleiner Perkins Christmas party, Mr. Singh and Mr. Perkins met Mr. Welch, who was then an executive vice president at the optical-component maker JDS Uniphase. Mr. Welch had been the chief technology officer of SDL, another Silicon Valley optical-component company, which JDS Uniphase snapped up for $41 billion. Mr. Singh and Mr. Perkins, meanwhile, were toiling at their metropolitan area network (MAN) services startup, OnFiber Communications. In 1999, the duo had sold Lightera Networks, a maker of MAN equipment, to Ciena for $625 million.

Amid cocktails and holiday cheer, the conversation turned to how economically unfeasible optical networks were. The three men found that they shared similar ideas about how to fix the network economics, namely by reducing the cost of network equipment. They made promises to meet soon, and all went their merry ways.

Back at OnFiber, Mr. Singh and Mr. Perkins found themselves growing restless. The itch to start Infinera perhaps struck Mr. Singh first. At 34, he is the youngest son of an Indian diplomat and grew up in India, Sierra Leone, and Jamaica before his family settled in Washington, D.C. OnFiber was Mr. Singh’s third company (Mr. Singh was one of Red Herring’s Top Ten Entrepreneurs for 2001).

In contrast, Mr. Perkins, 38, was born in Yonkers, New York, and grew up in Lancaster, Pennsylvania. In 1989, when Mr. Singh was still in college, Mr. Perkins was helping develop the point-to-point protocol, which allows two computers to communicate with each other. This is the cornerstone technology of the Internet–it allows a PC to connect to servers over a phone line.

Mr. Singh and Mr. Perkins, excited by their conversation with Mr. Welch at the holiday party, decided to start a company to build a new device that would make optical networks cheaper. But they soon realized that any hope of building this device hinged upon the development of a new chip–the integrated photonic circuit.

It was time to call Mr. Welch. Since 1988, Mr. Welch, one of the smartest minds in the world of optical components, has been granted 30 patents in the field of optics. While not exactly bedtime reading, his research papers, read chronologically, chart the history of the development of modern optical components. If Mr. Singh and Mr. Perkins were thinking about the box, Mr. Welch was thinking about its brain. Mr. Welch, however, could not commit to anything until May 2001, when his contract with JDS Uniphase expired. So they waited.

In short order, Mr. Singh and Mr. Perkins hired replacements for themselves at OnFiber, and Mr. Welch prepared to leave JDS Uniphase. On May 1, the trio formed Infinera (which until March operated in stealth mode as Zepton Networks). It became obvious that they would have to do it all simultaneously–design the chip, determine the best manufacturing technique to mass-produce it, and then build a box to showcase the power of their integrated photonic circuit–a daunting task. “In order to build a permanent company, you need to have a big problem to solve,” admits Mr. Singh.

Aside from tackling the problem of controlling light, Infinera’s chip also overcomes many of the limitations of silicon-based chips. Indium phosphide is one of what are called III-V compounds. It has photonic properties that allow for the integration of lasers, detectors, dynamic components, passive waveguides, and electronics. Chip makers are thus able to design a photonic circuit and etch it onto an InP wafer in much the same way as with silicon. And like silicon, InP wafers, which are typically three inches in diameter, can hold many microscopic circuits.

A Beautiful Chip

In December, Infinera began manufacturing test batches of its integrated photonic circuit from an in-house minifoundry. Though Infinera permitted Red Herring to visit its facilities and to speak with its engineers and executives, no one would reveal how many network functions lie within its integrated photonic circuit. We saw one such circuit with our own eyes (through a microscope of course) and thought the complex patterns aesthetically beautiful. Not being chip engineers, we could only play art critic.

Engineers, however, would be intrigued by Infinera’s ability to overcome what the world of electronics calls the “tyranny of numbers.” In the ’50s, complex digital systems were built from discrete electrical components like capacitors, resistors, and transistors. On the tenth anniversary of the birth of the first electronic transistor, J. A. Morton, a chip engineer at Bell Labs, wrote: “Each element must be made, tested, packed, shipped, unpacked, retested, and interconnected one at a time to produce a whole system. Each element and its connections must operate reliably if the system is to function as a whole . . . the tyranny of large systems sets up a numbers barrier to future advances if we must rely on individual discrete components for producing large systems.”

The electronics industry first overcame the tyranny of numbers in 1958, when Jack Kilby, then an engineer at Texas Instruments, came up with the first working prototype of an integrated system built from silicon. Around the same time, Gordon Moore, one of the cofounders of Intel, built a similar circuit. That integrated circuit (IC) became the cornerstone of the electronics revolution. While Mr. Moore’s breakthrough ultimately led Intel to its Pentium processors, Mr. Kilby’s approach, dubbed “the monolithic idea,” won him a Nobel Prize in Physics in 2000.

Forty-four years later, the optical communications industry is dogged by a similar problem. Optical-networking gear uses components that are manufactured and assembled one at a time, making most equipment very expensive. Components account for as much as 50 percent of the cost of a $100,000-plus long-haul DWDM device and 15 percent of a $40,000 piece of MAN equipment.

The tyranny of numbers has been the undoing of the likes of Corning and JDS Uniphase, two of the largest components manufacturers. Corning, which at one time was worth around $115 a share, now trades at $8 (NYSE: GLW). JDS Uniphase traded for more than $150 a share in early 2000 and now is worth only about $7 (Nasdaq: JDSU). Together, they laid off more than 20,000 people in 2001.

Meanwhile, once-strong demand from equipment makers like Cisco Systems, Ciena, and Nortel Networks has stalled and shows little sign of rebound. It doesn’t matter how many ways components slice a ray of light or how they transmit data through fiber; if they are too expensive, demand will remain soft.

If components could be integrated, demand for the chips powering them could leap from $14 million in 2000 to $2.6 billion in 2005, according to the Eastern Management Group, a telecommunications consultancy. This growth ultimately depends on the ability of manufacturers to integrate multiple network components, says David Yedwab, an executive vice president at the Eastern Management Group. The resulting cost savings in network construction and maintenance “could be huge,” he says.

Some of Infinera’s 700 or so competitors believe that the best way to create such an integrated component is by using planar light wave circuits (PLCs) instead of an integrated photonic circuit. PLCs are formed on glass or silica films. The approach among these companies involves using PLCs alongside conventional electronic chips to reduce network costs.

Genoa, in Fremont, California, is pursuing this method. The company has managed to shrink an optical amplifier, a component typically the size of a videocassette, down to the size of a sugar cube. Normally optical amplifiers cost between $5,000 and $30,000 each, but Genoa’s technology can reduce those prices by as much as 90 percent. The company’s ultimate goal, however, is a complete IC–which Infinera has done.

Other competitors in the race include CyOptics, Opto Speed, and Qusion Technologies. These companies are building components that perform a discrete function, like amplification, on a single chip. Meanwhile, Infinera is building a complete system on a chip. Using the fast-food industry as metaphor, many of these startups are building buns, patties, and pickles, but Infinera is building the Big Mac.

Of course, Infinera’s ultimate challenge is finding hungry customers among the ailing telecommunications companies. “There is a huge difference between coming up with an integrated component and getting Verizon to use it,” warns Mr. Yedwab.

Mr. Singh admits he worries about this. “Look, this represents the first optical integrated circuit,” he says. “And we do not underestimate what it takes to go from proof of concept to shipping product. It will be some time before we can perfect the process to produce and package these devices, and it will be more than 12 months–perhaps in 2003, when we go out of the door.”

Waiting until next year gives Infinera time to perfect a manufacturing process. So far no one has been able to produce InP-based circuits efficiently, but the company recently hired Fred Kish as its vice president of manufacturing. Formerly with the fiber-optic group at the equipment maker Agilent Technologies, Mr. Kish holds 28 patents, most related to chip processes. He will work with Mr. Welch on building a foundry.

Absolutely Fab

“Our biggest challenge right now is to get our fab up and running; we want to own the process to finish the product,” says Mr. Singh. In January, the firm relocated to Sunnyvale, California, and a month later it broke ground there on its chip foundry, or fab.

Infinera’s success depends on whether its bottom-up approach to lowering network costs works. But rather than wait for its inexpensive integrated photonic chip to to be used in devices on optical networks, Infinera is building its own network device based on its chip. Mr. Singh predicts that such a device could cut by half the cost of building, owning, and managing an optical network. If it costs $600 million to build and run a fiber-optic link between New York and San Francisco using current technology, he says, then Infinera’s technology could cut the costs to $300 million. But is that good enough?

“Half is not good enough; you need to do more than that. Anything that cost a buck before will have to cost a dime,” says Mr. Yedwab.

Infinera is heeding this message. Even though Mr. Singh, Mr. Perkins, and Mr. Welch have developed a chip that might cut network costs in half, they know that isn’t enough. They continue to look for additional ways to make optical communications even cheaper through applications of their integrated photonic circuit. If successful, their effort promises to rescue the telecommunications industry and to herald a new era of innovation.

© 1993-2005 Red Herring, Inc. All rights reserved.