Processor fabrication technology will change dramatically in 15 to
20 years. But until then, companies will have to squeeze more and
more from their silicon. Danny Bradbury reports.
Processing power is everything in today's computing environment.
The likes of Intel and Sun Microsystems invest millions of dollars
in research and development to make their chips faster than their
rivals'.
And thanks to their efforts, Moore's Law, the prediction made by
Intel founder Gordon Moore, has been proved correct. In 1965, Moore
commented that the capacity of memory chips doubled every 18 to 24
months. This has proved consistent in the processor world, where
the number of transistors on a chip has grown exponentially with
each generation.
One of the biggest issues for processor manufacturers is the
difficulty of fabricating smaller components on a chip. The size of
the processor is limited by the demand for smaller computing
devices, which is greatest in the consumer-computing sector.
Consequently, to fit more transistors onto a single processor, they
must be smaller and so must the connections. This is easier said
than done.
Processors are produced by a technique similar to photography
called lithography. Manufacturers produce an image of the
electronic circuit that they want to fit onto the silicon chip.
They then shine a light through the image onto photosensitive
material on the surface of the silicon. The resulting image is then
etched into the surface of the silicon. This is done several times
until a multilayer grid of electronic circuitry is built up on the
surface of the chip.
Producing smaller components involves expensive refinements to
the process and scientists also have to contend with the laws of
physics, which make it difficult to produce components smaller than
the wavelength of light used to make the image in the first
place.
Consequently, many manufacturers will tell you that the cost of
fabrication is increasing dramatically, leading to a consolidation
in the market. This is certainly the view of the California-based
PriceWaterhouseCoopers (PWC) Technology Forecast. The report, which
includes a section dedicated to processor technology, says more
companies will find it uneconomical to operate their own facilities
and will outsource fabrication to specialists.
Malcolm Penn, managing director of specialist semiconductor
market research company Future Horizons, says,"In absolute terms,
fabrication is more expensive, but relative to company revenues we
spend the same. It's no more expensive to build a fabrication plant
now than it was 30 years ago."
Furthermore, we are unlikely to see a quantum leap in processor
fabrication methods within the next two decades. Many new methods
of processing have been proposed that depart drastically from
conventional methods. But firms are doing everything to squeeze
more functionality out of existing processes and they will be able
to do this for some time. The PWC report says lithographic
fabrication will continue until at least 2014, while Penn predicts
we won't see advanced methods used on a commercial basis until
2050.
Meanwhile, companies will have to be content with enhancing
current methods of fabrication and making changes to the logic
control software hard-wired into the processors. Firms have already
moved from visible to ultraviolet light, which has a shorter
wavelength, as a means of producing more detailed electronic
circuitry on a processor. Other light sources such as extreme
ultraviolet, X-ray lithography and electron beam lithography are
also being explored.
Extreme ultraviolet could produce line widths much smaller than
those available today and they could make the chips up to 100 times
faster, according to the PWC report. X-ray lithography has been
successfully demonstrated by IBM, while electron beam projection
uses electrons instead of photons in the lithography process.
Firms are also using new materials to create faster processors.
There are three well-known technologies that have already produced
commercial products:
- Silicon on insulator - processors are slowed by the capacitance
(the ability to retain an electrical charge) of the transistor. The
greater the capacitance, the longer it takes to alter the charge
within a transistor and open or close the electrical gate that the
transistor creates. Insulating the transistor from the silicon
substrate reduces the capacitance and enables it to switch
faster.
- Copper - this has a much greater conductivity than aluminum,
which is traditionally used to hook together transistors on a
silicon chip. Because copper conducts electricity more efficiently,
electrons travel along it faster, leading to a faster processor.
IBM has been making great strides in this area.
- Silicon germanium - this uses conventional silicon as the
substrate for a processor, but employs silicon mixed with germanium
as the basis for the transistor material. Germanium can operate at
a lower semiconductor bandgap - the difference in charge between an
"on" and "off" device. Consequently, silicon germanium chips can be
faster and/or more power efficient. It is becoming a mature
technology in the commercial sector, following IBM's commercial
introduction of devices based on the technology in 1998.
The other way of creating faster processors is to use more
intelligent hard-wired software on the processor, enabling it to
handle instructions more effectively. The most recent development
in this area is explicitly parallel instruction computing (Epic).
This extends an existing feature of modern processor design, in
which the processor carries out multiple instructions at once, by
using different pipelines down which instructions can be sent.
Unlike previous parallel instruction processing architectures,
an Epic-based architecture can add more pipelines in subsequent
generations without rendering itself incompatible with existing
software, according to the PWC report.
The first Epic-based processor family is the IA-64, a result of
joint work between Intel and HP. The first processor in this
series, called Itanium, was supposed to have shipped last year but
it has been delayed. It will be followed by McKinley at the end of
this year. Like Itanium, McKinley will be targeted at the high-end
server market.
In the short to mid-term, the future for processing technology
will focus on producing more densely-packed silicon surface areas.
Companies will use enhancements to existing fabrication
technologies to create more complex chips with greater numbers of
transistors, and technologies such as Epic will result in more
intelligent instruction processing without losing software
compatibility.
The next quantum leap in processing technology is unlikely to
happen for at least 10 to 15 years, at which point we will have
exhausted the potential for improving lithographic technology.
Then, the transition from painstaking enhancements to dynamic
exploration will begin.
Goodbye to Intel?
Fabrication processes: how small could you go?
While conventional fabrication processes are likely to be around
for the next decade at least, the technology to support future
processors is already in the labs. Molecular computing involves the
manipulation of logical components created from atoms.
Hewlett-Packard has been heavily involved in this research and, in
July 1999, scientists at the company's laboratory managed to
engineer a molecular-level logic gate. This would do away with
lithography altogether.
With some imagination, it is easy to think of numerous
innovative applications for molecular computing. Because you are
dealing with another form of nano-technology, it is theoretically
possible to build a supercomputer into your nail polish or tattoo a
mainframe onto your body. Injecting tiny computers into your body
to locate, analyse and neutralise cancers or viruses could be an
option.
But let's not get ahead of ourselves. Philip Kuekes, a research
scientist at HP labs, explains there are many obstacles to be
overcome before molecular computing becomes a commercial reality.
For one thing, getting information to flow through the matrix of
molecules is no mean feat.
"The limiting aspect of almost all these technologies is not how
fast you can sense or flip a bit," Kuekes says. "It's how fast you
can get the information to the outside world if the bit has one
million neighbours. You have to think about signaling down those
wires."
Quantum computing is even more complex than molecular computing
because it makes use of the Superposition Principle, which only
works at quantum level. This principle states that some quantum
particles only exist in a certain state when observed - if they are
not observed, then they can exist in multiple states at once.
Because the state of a bit represents a number in computing, this
makes it possible to perform calculations on different numbers at
the same time.
Quantum computing has huge potential and could change life as we
know it. But it is also going to take decades before, or indeed if,
it becomes commercially available.
Finally, the last Star Trek-like processor technology is DNA
computing. Scientists have proposed that, because DNA is already
used to store information, it could store and process information
put there by programs.
How fast chips could help in daily life
Rodney is unhappy. He is in a foreign country on a business trip
where he cannot speak the language and his multilingual assistant
has been taken ill. Also he has to process an immense amount of
marketing information before he can make a presentation to a
potential client. Luckily, Rodney has a personal digital assistant
with a low-power, high-speed processor that has enough capability
to process language information. Trying to negotiate with a hotel
concierge, Rodney holds up the palmtop, which listens to the
concierge's message, translates it and relays it to Rodney in his
language.
But he still has to process those sales figures, which involves
analysing the telephone call statistics of hundreds of thousands of
telecommunications customers, and then cross-referencing them
against customer salary information. There are just five minutes
before his presentation is due to start. He calls the office and
asks a colleague to run the numbers. Until a couple of years ago,
it would have taken hours to produce the results. Now, it takes a
few seconds because the quantum computer sitting on his colleague's
desk uses quantum uncertainty theory. This puts computer bits in
two states at once - both on and off - so calculations can be made
on multiple numbers at the same time. Strange, but true. And very
useful for Rodney.