Skip to Content

World’s Fastest Optical Chip

How Infinera packs dozens of optical components onto photonic integrated circuits for ultrafast optical networks.
January 1, 2007

In his lab in Sunnyvale, CA, ­David Welch, cofounder of telecom startup Infinera, holds up a rigid two-­centimeter-wide strip featuring four patterned, gold-colored rectangles. It’s made of indium phosphide, a semiconductor prized for its optical properties. The chip’s simple appearance belies its complex engineering and gives little hint that it could be the key to cheaply supplying the bandwidth demanded by a YouTube-addicted world.

Shown here are fourteen 100-gigabit photonic integrated circuits sitting in a plastic carrier for performance testing.

The gadget is called a photonic integrated circuit, and it represents an important practical advance in optical data transmission. Since the early 1990s, such transmission has increasingly relied on a technique called wavelength division multiplexing (WDM). With WDM, data is encoded on as many as 80 laser beams, each having a different wavelength. Those beams are then combined for a trip down an optical fiber thinner than a human hair. At a node on the other end of the fiber, the beams are split into their constituent wavelengths, and the information is turned into the electrical signals that reach our computers.

The optical equipment required to do all this includes lasers that send light, multiplexers that split it up or recombine it, modulators that encode it with data, and detectors that receive it. Traditionally, these devices have been housed in their own little packages, each about the size of a pack of gum, and combinations of them were bulky, expensive, and sometimes unreliable. Infinera–founded in 2001 by veteran executives and technologists from optical­-telecom leaders like Ciena and JDS Uniphase–set out to put dozens of such components on a chip, the way electrical engineers combine transistors in an electronic integrated circuit. “What nobody had tried to do was essentially put an entire WDM system on a pair of chips [one to send, the other to receive], and nobody had tried to commercially manufacture it,” says Welch. Infinera not only tried to do both but succeeded.

In 2004 the company introduced the first large-scale photonic integrated circuit–a chip with 50 nanoscale optical components patterned into its surface. Previously, other optical-chip manufacturers had managed to integrate only a few such devices on a single chip. The first Infinera device was capable of sending or receiving 100 gigabits of information per second. Now the company has demonstrated a 400-gigabit chip and is well along in the development of what it describes as the fastest optical chip in the world–a 1.6-terabit version that it expects to commercialize within several years. The four gold patches on the chip in Welch’s hand contain an astonishing total of 240 patterned optical components.

Multimedia

  • A demo of how Infinera makes ultrafast optical networks

Of course, despite the theoretical advantages of an “all-optical Internet,” no network is based entirely on optics. Equipment at network nodes converts optical signals to electrical ones so it can clean them up and amplify them, or deliver them to a computer. Infinera’s technology does this, too, passing some jobs off to microprocessors on a circuit board that will then transfer them back.

But the photonic integrated circuit reduced the cost and complexity of the conversion process. This advantage, in turn, allowed Infinera to promote a new network architecture–essentially, one with more network nodes. Other companies had tried to keep costs down by reducing the number of nodes, with their traditionally bulky optical devices.

Having more nodes means more flexibility to add access points and easier maintenance and fault detection. It thus makes it easier to combine the benefits of optics and electronics. And the Infinera package–chips and circuit boards–take up one-fifth the space of conventional technology.

Late last year the Internet2 consortium–a group of more than 300 U.S. government, university, and corporate research centers that need high bandwidth to share everything from particle-physics data to medical images–began deploying a new optical network that uses Infinera’s systems. “Infinera’s technology is unique,” says Steve Cotter, director of network services at Internet2. “Instead of trying to avoid optical-electrical transitions, they made them cost effective.”

Photonic Fabrication

Making the Infinera chips is no simple task. Optical devices are three-­dimensional structures, far more challenging to manufacture than two-dimensional silicon transistors. Making the lasers, detectors, modulators, and other components of the finished chip requires repeatedly depositing and etching away many thin layers of different materials, such as indium gallium arsenide and indium phosphide.

Infinera’s process starts with a wafer of indium phosphide. The wafer moves along an assembly line, where it is coated with a syrupy chemical called photoresist. Ultraviolet light shines through a mask with stencil-like designs and irradiates the photoresist, effectively “developing” intricate patterns that allow some semiconductor material to stay on the wafer and some to be etched away.

At a high level, it’s the same as the photolithography that companies like Intel use to make silicon microprocessors for your PC. But there’s an important difference. “In an Intel chip, it’s all silicon. In optics you use various semiconductors with various functions,” Welch says. And the indium phosphide wafers go through many more rounds of deposition and etching than silicon wafers do. Infinera is tight-lipped about the details of its manufacturing process, which was designed with the help of engineers experienced in such tasks as manufacturing silicon microchips and mass-producing light-emitting diodes. Welch says the company has exclusive patents on key aspects of the technology for placing large numbers of devices on indium phosphide wafers.

The 1.6-terabit chip differs from the 100-gigabit version largely in the number of devices patterned onto it. Each 100-gigabit chips contains, among other components, 10 lasers, 10 detectors, 10 modulators (which encode data by switching light on and off), and 10 waveguides that direct photons into a multiplexer. The 1.6-terabit chip’s 240 components include 40 lasers, 40 detectors, 40 modulators, and 40 channels. And each modulator encodes data four times as fast.

After the wafers come off the line, they are sliced into chips–several hundred of them. Finally, the chips are tested for potential malfunctions, combined with electronic chips built by Infinera on a device called a line card, and installed in optical networking units for shipment.

Demand for Internet video and voice services is exploding, threatening to overwhelm the typical broadband connection, which transmits between one and six megabits per second. “We’re all thinking that people will need 25, 50, or 100 megabits,” Welch says. To meet that demand, Internet companies will have to pack more equipment into already overcrowded switching stations. “With Internet traffic growing at 60 to 100 percent per year, you can’t keep installing ­refrigerator-size racks in the basement,” Welch says. “Photonic integration becomes the technology that enables the Internet to grow.”

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

Google’s Gemini is now in everything. Here’s how you can try it out.

Gmail, Docs, and more will now come with Gemini baked in. But Europeans will have to wait before they can download the app.

This baby with a head camera helped teach an AI how kids learn language

A neural network trained on the experiences of a single young child managed to learn one of the core components of language: how to match words to the objects they represent.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.