|
University of California Santa Barbara Professor John Bowers and Intel Fellow and 2008 R&D Magazine Scientist of the Year Mario Paniccia collaborated on the 2006 invention of the hybrid silicon laser.
|
Fiber
optics are a staple of modern communications life. With it we have
high-definition video, rapid data transfer, and (mostly) painless
Internet browsing.
But
we’re still running out of room. As some of us already know,
high-definition movies are a chore to download. In the workplace, files
sizes have become so huge that simply sharing them or moving them around
has become a chore. In imaging, petabyte file collections are now
commonplace. In high-energy physics, researchers are well in the
exabytes-per-year class.
On Tuesday, R&D Magazine Scientists
of the Year Mario Paniccia (2008) and Justin Rattner (1989)
demonstrated what they believe will help break us free of some of the
communication constraints that are hindering both consumers and
researchers.
The
breakthrough is a silicon optics link that represents the first major
practical application for hybrid silicon technology that was unveiled in
2006 and was the product of research at Paniccia’s lab Intel Corp.
Paniccia, an Intel Fellow, has been making headlines with his
laboratory’s breakthroughs, the most recent of which was a fast
avalanche photodetector based on silicon.
But
in a demonstration on Tuesday, he and Rattner, Intel’s CTO and a Senior
Fellow, showed that the technology really does work. They sent data
seamlessly from Santa Clara to Monterey, Calif., at a blistering 50
Gbps.
“This
was done in air, with no special cooling. It achieved a petabit rate
with almost zero errors. It’s equivalent to 3x10-15, which is pretty
good. It gives you an idea of the robustness of the platform,” says
Paniccia. The result is that a typical HD movie can be transferred in
about 30 seconds. He pointed out that the link is still a prototype and
that more work will need to be done before a product can be delivered.
|
Rated for 50 Gbps, the prototype silicon optical link can transmit an average HD movie in about 30 seconds. Intel says the link is designed to be scalable to higher speeds, and Paniccia has a target of 1 Tbps, fast enough to download a movie in a fraction of a second.
|
“The
next step is take it to high-volume manufacturing, and we’ve proven we
can do it,” he says. Earlier, the link was operated for more than 24
hours, achieving similar performance.
In 2008, when R&D Magazine
interviewed scientist Mario Paniccia about his unconventional research
at Intel Corp., he said the goal for the company was to eventually
produce an integrated chip, from silicon, that would do everything a
conventional optical photonics link could do.
But
Paniccia, and Intel, have their sights set on a bigger goal: the
transformation of the industry itself. The optical link is just a small
piece of the effort to lose the copper wires that still dominate
personal computing and bring fiber optics to everyone.
As
for Paniccia’s lab, he has three phases in mind: the first is to prove
that the pieces can be built; the second is to integrate the pieces; and
the third is to produce it using low-cost assembly technique and
high-volume manufacturing practices. Essentially, Intel needs to ensure
that the product is both affordable and profitable to produce.
“The
milestone today is the next phase, the integrated silicon photonics
chip. But it’s just the beginning. 50 Gbps is today, but we have a path
that will bring us to 400, 500, eventually 1 Tbps. We have put enough
pieces together that we think we can scale the technology,” he says.
The company has taken other steps to help make this happen by introducing a “Light Leak”
high-speed
optical cable technology designed to connect electronic devices to each
other at speeds from 10 Gbps to eventually 100 Gbps over the next
decade. If the optical links and personal computers are built to match,
that HD movie will arrive in less than second.
Simplified with silicon
Paniccia’s
invention is designed to emit, manipulate, combine, separate and detect
light. It does all this in much the same fashion as a conventional
fiber optics communications link. The optical information is received on
several channels (in the case of the prototype, 4), which are combined
by way of multiplexing and sent through the link to a demultiplexer,
where the signals are once again separated and converted to electricity
by a photodetector.
|
The hybrid silicon laser used in the 50 Gbps optical link introduced to the public on Tuesday relies on a layer of indium phosphide. When fed electricity, the material emits photons, which are collected and amplified by a cavity etched in silicon. This produces optical signals suitable for carrying data.
|
In
the case of Intel’s silicon optical link, however, this is done in
miniature on a single chip. The key component is the laser. In 2006,
Paniccia and Prof. John Bowers of the University of California Santa
Barbara developed a unique process to fuse indium phosphide to silicon.
The InP layer emits light, which is collected and channeled by a
waveguide etched directly into the silicon chip. Though it was not
widely believed at first that silicon could act as a laser cavity, it
was demonstrated by Intel and led to a key development in 2008: etched
gratings were added to the waveguides, creating lots of tiny mirrors
than could be used to create multiple wavelengths of light. In short,
Intel had a tunable laser-in-a-chip that could be created by the
thousands with a single bond.
Paniccia
was quick to point out the relative irony of the announcement being
made during the 50th anniversary of the laser. Despite the ubiquity and
usefulness of lasers, he says, they simply haven’t progressed the way
integrated circuitry has moved from a single transistor to trillions on a
single wafer.
“Today,
lasers are used everywhere and the technology for long-haul
communications is everywhere. The problem is that you can’t use them
anywhere around the computer,” says Paniccia, because they are
complicated and expensive. Worse, the assemblies required to control the
way lasers combine and split signals (multiplexing and demultiplexing)
between fiber strands require expensive materials, optics and thermal
control strategies.
To
build the link, the 4-channel silicon laser is installed in a
transmitter chip. The signals from the lasers pass through silicon
modulators, another Intel invention that was introduced in 1 GHz form in
2004 by Paniccia’s lab and has since been accelerated to 40 Gbps as of
2007. These modulator prepare the signal for multiplexing--on a tiny
scale--and eventual transmission to the receiver chip.
Another
2007 innovation from the lab, germanium-on-silicon photodetectors, have
been further improved and are used to convert the optical signal to
electricity after demultiplexing. The transmitter and receiver both
incorporate V-channel pins that allow each passive optical connection to
be attached and detached quickly.
The
parallel channels in the laser, says Paniccia, are the key to scaling
this technology. He anticipates eventually building 8-channel chips, or
even stacking them vertically to multiply the data flow. First, though,
the costs must be carefully controlled.
“Early
in the process we realized we had to address the packaging assembly. So
we’ve run silicon photonics down the same road as other technologies,
like the integrated circuit,” he says.
That
meant configuring everything to be installed on an average low-cost
printed circuit board, the use of inexpensive flip-chip bonding that is
commonplace to high-throughput chip-building, and passive connections.
Possibilities on the horizon
Data
speed is the easiest way to envision the potential of this technology,
but the true test will be how effectively the pieces that Paniccia’s lab
has built--and continue to invent--will fit into what is a carefully
constructed and highly competitive information technology space.
Paniccia
and his team have given thought to that, but he say it’s too early to
really say for sure where the first such links will appear. Undoubtedly,
he says, it will affect the way personal computers send and receive
information.
“When
we talk about applying this anywhere and everywhere, we’re talking
about in the server space, in the data center, and CPU to CPU. This is
really about communications,” says Paniccia.
But
that doesn’t preclude other ways to use optical links. In PCs, for
example, one of the biggeset limitations is memory. The average computer
only has 4 DIMM slot because of electrical limitations. It’s
conceivable with silicon optics that the memory chips can located in a
server rack completely separate from the motherboard.
“We
could completely re-architect and install more memory than anyone
considered possible,” he says. On the client side, high-definition video
is partly limited by the need for HDMI cables. But using optical links,
the bottleneck is gone.
The
most exciting application for Paniccia, however, is in biotechnology.
He says these links fulfill all the criteria for biosensors and
biochips, and represent a potential building block for biometrics
solutions of the future.
These
solutions are still far in the future, Paniccia says, but his target
isn’t so much the application as it is the rate of improvement. Tied as
Intel to the tenets of Moore’s Law, which predicts a rise in
microprocessor performance relative to a decrease in its cost, the
mission is to build a technology platform that drives applications for
silicon photonics in an unceasing stream.
To
that end, the company has done pricing and cost analyses, and run these
new modules through pilot lines. Soon the technology will pass into the
hands of the high-volume manufacturing experts at Intel who will solve
new problems.
“It’s
really about driving to commercialization, hopefully by the middle of
this decade. We’re on an aggressive path. The goal is to take high
bandwidth everywhere and to integrate it over time,” says Paniccia.
Read more at Intel
Conductor of Light: Mario Paniccia, 2008 R&D Magazine Scientist of the Year
R&D Magazine's Scientist of the Year Awards Program
