IEEE Spectrum September, 2010 - 35

VCSEL: A vertical-cavity surface-emitting laser
emits rays perpendicular to the chip.

PD: A photodetector receives
signals from the laser.

LDD: A laser diode
driver converts an
electrical input signal
into a modulated
current and supplies
it to the laser.

POLYMER
WAVEGUIDES:
Light-guiding optical
channels are laid
down directly on the
circuit board.

CHIP CARRIER:
Provides electrical
input/output.
OPTOCHIP LENS
ARRAY: Collects light
from VCSELs and
focuses light into
photodetectors.

paths in the vias, the vertical conductors
that connect elements of a circuit board.
It gets worse still as bit rates approach
10 gigabits per second, when cross talk
blurs the signal, even at distances of less
than a meter.
These problems are particularly bad
when you're yoking together the multichip modules of a massively parallel
computer. When one module must link
up with another at the other end of a circuit board or, worse still, in a different
rack of equipment, the bandwidth bottleneck becomes particularly severe. That's
why today's highly parallel machines
can reach peak performance only when
solving those specialized problems that
can be readily divided into many tasks
that can be processed independently.
By avoiding all those signal-loss and
cross-talk problems, an optical bus
would make supercomputers go much
faster. It would also make them easier to program, because programmers
wouldn't have to take special measures
to compensate for such severe communication delays among processors.
spectrum.ieee.org

TIA: A transimpedance ampliﬁer
strengthens the
electrical current from
the photodetector.

Fiber-opTic lines FirsT began proliferating in the 1980s in long-distance
telecom networks. By the late 1990s,
fiber-optic links had found their way
into local and storage area networks,
interconnecting systems hundreds
of meters apart. Over the next decade,
the technology kept moving down to
ever smaller dimensions as its cost and
power needs kept falling and the bandwidth requirements of computer systems kept rising.
Consider the role of optical interconnects in supercomputing. In 2003
and 2004 , the fastest machines on
Earth were the NEC Earth Simulator
and IBM's initial Blue Gene L, which
had peak performances of around
36 trillion f loating-point operations
per second, or teraflops. All the interconnects in those supercomputers
were electrical. Then, in 2008, IBM's
Roadrunner set a new record by achieving a quadrillion (a thousand million
million) such operations per second,
which is known as a petaf lop. The
Roadrunner used some 40 000 optical

WAVEGUIDE
LENS ARRAY:
Conveys and
focuses light into
and out of polymer
waveguides.

links to connect entire racks of servers.
Sometime next year, at the National
Center for Supercomputing Applications
at the University of Illinois at UrbanaChampaign, IBM's Blue Waters machine
is expected to go 10 times as fast, using
more than 1 million optical interconnects.
The number of parallel optical modules
inside Blue Waters will be roughly equal
to the world's current annual production. IBM will be a major consumer of
these optical interconnects, and though
it helped to pioneer parallel optics, IBM
does not plan to go into the business of
manufacturing the modules. So the company is sharing much of its knowledge
with potential suppliers.
Computers beyond Blue Waters will
need proportionately more interconnects. Connect the dots and the resulting trend suggests that 1000-petaflop
(Pf lop) systems will require an astonishing 400 million optical links. In general, judging from past performance data,
supercomputer processing rates have
been going up tenfold every four years.
So you can expect
Continued on page 52
SEptEmbEr 2010 * iEEE SpEctrum * NA

http://spectrum.ieee.org

Table of Contents for the Digital Edition of IEEE Spectrum September, 2010