There was once a country that had done away with that scourge
of modern living: urban gridlock. The leaders of this faraway place built
multi-lane highways through their cities, abolished traffic lights and thus
managed to boost average speeds to over 100 kmph. There was only one problem:
between the cities, there were only bumpy dirt roads; whenever vehicles ventured
beyond the beltways, they were forced to slow to a snail’s pace. Despite the
urban road network–and notwithstanding the economic activity this encouraged–the
country as a whole remained underdeveloped.
So the leaders reconvened and decided that the way to promote
trade and economic activity between the various regions was to build–not an
interstate highway, but more dirt roads. If you think this country doesn’t
exist, you’re right. But translated to the world of IT, it’s a situation
that’s unfortunately true. Each and every computer approximates to such a
country: the processor and memory chips on the printed circuit board (PCB) are
the high-tech cities of the parable, the connections between them the bumpy dirt
roads.
"The electrical connections are the real bottleneck of
computer technology," explains Elmar Griese, an electronics engineer at
Siemens in Paderborn, Germany. "The top processors today can crunch between
one and two billion bits–i.e. units of information–a second, but the signals
between the components only travel at about a tenth of that speed."
The result is very similar to the situation in our fictitious
country: "the processor sits around waiting for data to come from the
memory or other microchips. In other words, no computer ever really works as
fast as it could." Griese is currently studying ways to remedying this
problem. His team of specialists in the field of optical connection technology
belongs to C-LAB, a collaborative venture between Siemens and the University of
Paderborn. At present, 30 people from Siemens and 30 from the university are
looking at the technologies of the future–projects ranging from Internet for
the blind to robots that can be controlled via the Net.
In 1998, C-LAB got together with partners from industry and
research to launch a unique project. The aim was to speed up data transfer
within the computer itself. Today, chip manufacturers still resort to a
makeshift solution in order to handle the growing volume of data throughput–they
simply multiply the number of entrance and exit points to the chips, i.e. they
build more dirt roads. "That can’t be the ultimate answer," says
Griese, "If we really want to find a long-term solution, we’re going to
have to turn our back on electronics."
The use of electrical connections has a number of
disadvantages. First of all, it offers high electrical resistance to rapidly
alternating signals and thereby alters the shape of the pulses. As a result,
electrical connections can transmit 100 million bits per second, but not the one
or two billion that the best processors can process in the same time.
Moreover, electrical connections emit, and are affected by,
electromagnetic radiation. In essence, they function like antennas. If
individual connections are laid too close, they interfere with one another, so a
minimum separation must be maintained.
Nevertheless, such problems are avoidable. As Griese
explains, the secret is to use light instead of electrons to transmit the data–"at
least in areas where you need high transfer rates." His team has been
looking into the use of very thin, transparent rails made of plastic. These
so-called optical waveguides are directly integrated within the PCB itself.
While the microchips continue to work electrically, the data is transmitted
between them by means of pulses of light.
At the points where the data leaves the chips, tiny laser
diodes convert the electrical signals into pulses of infrared light. This is
then directed to a deeper level within the PCB, where an optical waveguide
speeds it on its way. It’s just as if the cars in our city were driven to its
outskirts, loaded onto high-speed trains, transported to the next city and then
unloaded again. In the world of computer hardware, the rail terminals correspond
to the optoelectric components used to convert electrical signals into pulses of
light and vice versa.
As always in the high-tech world, things are not quite so
simple in practice. The optical waveguides, for example, are only 0.1 mm (0.004
inches) thick–about the same size as a human hair–and are covered with an
even thinner coating. The latter is what reflects the light in a zigzag course
along the inside of the waveguide. In order to thread a beam of light through
such as minuscule structure, the components involved must be positioned to
within an accuracy of 0.01 millimeters. That’s something normally beyond even
the most advanced of today’s PCB assembly technology. Nevertheless, with the
help of some sophisticated connector geometry, production engineers have managed
to achieve such accuracy with conventional component insertion machinery.
Production of the waveguides also involves a fair portion of
innovation. First of all, a die is used to stamp the optical paths into a layer
of plastic. This is then sandwiched together with the other–electrical–layers
of the circuit board. All the work is carried out by the project partners from
industry and research, C-LAB is responsible for design, computer simulation and
the test systems.
The team has already achieved data transfer rates of three
gigabits per second with the hybrid PCBS. That’s way above anything achievable
with the very best electrical connections and around 20 times as fast as
conventional circuit boards. And that’s just the beginning. "In a few
years time, we’ll be reaching speeds three times as fast," Griese
predicts. His idea is to make the light travel through the waveguides along a
straighter path–i.e. cut the amount of zigzag. "That’s possible if we
can slim the waveguides down to around 0.03 millimeters."
The problem is that today’s methods of PCB production can’t
deliver such precision. "And the techniques we use have to remain
compatible with conventional manufacturing methods, or costs will explode,"
Griese explains. Even so, the results already achieved are so promising that
computer manufacturers would like to start using the hybrid PCBS as soon as
possible.
The new technology also has many other applications.
"You can use it wherever you need fast data throughput," says Griese.
That includes the base stations in future generations of mobile communications
technology, where networks will have to transmit 200 times as such data per
second as they do today. Such rates will enable cell phone users to transmit
images or even take part in a videoconference. Another potential application is
in the nodal computers used in fiber-optic networks, where data throughput rates
are already as high as hundreds of billions of bits per second.
Here, engineers encounter exactly the opposite problem to the
one they face with computer chips. Here, as it were, the volume of traffic in
the city is so high that cars begin to bunch together. Unlike the case in our
fictitious country, however, the cities are connected up by massive highways
where the traffic can move pretty much as fast as it likes. With a fiber-optic
link, billions of bits of information can be transported every second. Under
laboratory conditions, Siemens engineers have already managed to post world
record data-transfer rates of 3,200 Gbps along a 40-km-long fiber-optic cable.
That’s the equivalent of 50 million telephone calls at once.
Yet the computers at the end of such superhighways still work
with conventional PCBS; the data flows from component to component via
electrical circuitry–which rapidly begins to choke on such high throughput. At
present, the light pulses transmitted by the fiber-optic link have to be
converted into electrical signals before the nodal computers can process the
data. The extra technology required to do this not only increases costs but also
leads to problems of space. What’s more, the electrical circuitry is, as ever,
prone to electromagnetic interference. "Our optoelectric PCBs will be just
the thing for these nodal computers, too," says Griese with a smile. His
vision of the ideal computer landscape is a place where the roads are excellent
both inside and outside the city limits: "That’s the only way to keep the
traffic moving."
Ulrich Eberl is a science journalist at Siemens Corporate
Communications. He lives in Munich.
Courtesy: New World, Siemens AG, Germany