Optical interconnects will become a game changer if we can combine the
high data rates, reach and integration of multiple photonic channels with
prices that are practical for system-level interconnects.
for a 3-D video conference in which the other party literally
appears to be in the room with you.
To keep the same quality and resolution, you might need
16 times the pixels of today’s HDTV. Transmitting such
“ultrahigh-definition” images at a 60-Hz refresh rate requires
60 Gbps, which is beyond what we can do with copper but is
easily accomplished with fiber optics. The problem is doing it
economically with fiber optics.
Phase II included solving issues of packaging, assembly and
fiber coupling, which contribute significantly to the cost of the
final fiber-optic components. Many 10 Gbps transceivers today
still require active, sometimes manual fiber alignment and
assembly procedures. The silicon photonics industry needs to
shift to fully automated, high-volume assembly techniques as
used in the PC industry; otherwise, the cost benefits of silicon
will be negated by pricy packaging.
Realizing silicon photonics
Before the turn of the millennium, silicon photonics remained
an unlikely candidate for fiber optics. Because silicon is optically
transparent at key optical communications wavelengths (around
1,310 and 1,550 nm), it could be used to create waveguides and
other passive devices but little else. Silicon lacked the necessary
physical properties for active devices—specifically, the direct
bandgap needed for light emission or the electro-optic (
Pock-els) effect typically used for the modulation of light.
Our goal has been to create active optical devices that can
be integrated and fabricated using the same manufacturing
equipment used to make silicon integrated circuits. Our vision
included three phases of research and development: 1) to prove
feasibility by demonstrating optical building blocks, then 2) to
move to integration, and finally 3) to implement high-volume
manufacturability. For the middle part of the past decade, we
focused on phase I, developing a variety of fundamental building blocks such as photodetectors, lasers and modulators, and
then scaling their performance to higher speeds.
It’s a testament to the progress that has been made in this
field to remember that, before 2004, such devices had not yet
been demonstrated at speeds greater than 20 Mbps
in silicon. Today speeds are measured in tens of gigabits per second, a viable silicon photonics industry
has emerged, and we have been able to demonstrate
not only 40 Gbps modulators, but 40 Gbps photodetectors, silicon Raman lasers and hybrid indium
phosphide-silicon lasers.
Multiplexer
The 50 Gbps silicon photonics link is the product of
that effort, and it required us to put devices with different process recipes onto a common substrate. While
in phase I, we enjoyed the freedom to optimize processing to produce individual devices. The integration phase required the team to tackle the practical
engineering challenges of consolidating process flows
and considering the thermal effects that devices such
as the lasers have upon neighboring devices. [ Key silicon photonic elements ] The link’s key silicon photonic elements, with the silicon photonic transmitter chip on the left and the silicon photonic receiver chip on the right. Courtesy of Intel Corp. Optical fiber Indium Phosphide layer Connectors Fiber Tapers Alignment pins
Soon after, we shifted focus from the development of building blocks to phase II— integration.
Modulators
Hybrid Si lasers
Photodetectors
Demultiplexer
Inside the 50 Gbps link
The 50 Gbps silicon photonics link is comprised of four basic
components: a silicon photonics transmitter (Tx) chip, a silicon
photonics receiver (Rx) chip, CMOS ICs (modulator driver,
transimpedance amplifier) chips and a passive fiber-optic
connector. These components are packaged and assembled on
low-cost printed circuit boards. The link uses coarse wavelength division multiplexing, with each chip containing four
optical channels operating at up to 12. 5 Gbps per channel for
an aggregate bandwidth of 50 Gbps on single fiber.
Hybrid silicon lasers
One of the most challenging but important technologies
in the link is also the basis of the transmitter—the aforementioned hybrid silicon laser. Developed by Intel and the
University of California at Santa Barbara, a hybrid silicon
laser consists of two elements: a silicon waveguide etched with
distributed-Bragg-grating mirrors to form the laser resonator
and an indium phosphide chip to act as a light emitter. The
chip is bonded to the silicon by passing an oxygen plasma over
the surface of both materials to form an oxide layer roughly
