¿ìè¶ÌÊÓƵ

Despite the stifling heat outside, at midday in Hiroko Sakaki’s office
in the University of Tokyo it is cool, dark and cramped. All the available
surfaces – the tables, shelves, chairs – are piled with papers, conference
proceedings, textbooks and novels in Japanese and English, and other paraphernalia
of academic life. Old-fashioned telephones with rotary dials sit on top
of yet more piles of paper. A vacuum cleaner sits, dusty and exhausted,
by a table. It looks as though nothing has been thrown away since the room
was first used in the 1950s. But it is here that Sakaki sits and reads and
thinks, producing ideas about quantum electronics which may make it possible
in the coming decade to build a new generation of computers, orders of magnitude
smaller and powerful than anything ever seen.

About fifteen years ago, Sakaki came up with an interesting idea: ‘quantum
wires’. If you could build a channel of semiconducting material just a few
atoms wide in a different semiconducting substrate, then an electron could
only run along it: provided the channel is deep enough and the temperature
low enough, the electron would not have enough thermal energy to jump out
of it. You would have built a one-dimensional wire. (See Figure 1)

Quantum wires could revolutionise computing. Electrons will travel much
faster because they will not bump into each other as they do in conventional
metal wires. Less power is needed; less heat is generated. That means the
computer’s circuits can be packed more densely and messages will reach their
destinations more rapidly. With quantum wires, everything could be linked
at the atomic scale; your supercomputer fits into a hardback, and uses
two penlight batteries. Would you like one for Christmas 2010?

Quantum leaps

But do quantum wires work? ‘It’s a real, material system,’ says Sakaki.
‘However, most of the experiments which give convincing evidence for quantum
wires have been at low temperatures.’ That’s because the channels built
so far have been hundreds of angstroms wide, rather than the 1 or 2 angstroms
that Sakaki imagined. At that size, the electrons in the ‘wires’ must be
drastically cooled to temperatures only a few degrees above absolute zero;
otherwise their thermal energy jolts them out of the channels.

But Sakaki is hopeful that more sophisticated fabrication techniques
will create quantum wires that function at room temperature. He points out
that as you shrink the channel size by a factor of two, the separation between
the electron’s energy levels goes down by a factor of four. ‘So at 1 angstrom
wide, it should work at 400 kelvin (127 °C), rather than 4 kelvin (-269
°°ä).’

The quantum approach is not the only way of making the next century’s
computers and memory chips smaller and faster. The other is optoelectronics,
which relies heavily on lasers and electronic elements that can convert
laser light to electrical impulses, and vice versa. Optoelectronics is perceived
as the coming technology for communications and for parallel computing.
Analysts in Tokyo predict that sales of optoelectronics devices in Japan
will be worth more than 32 000 billion yen (about £200 billion) by
2010, and Japanese companies are already world leaders in manufacturing
optoelectronics devices such as CD players and flat-panel displays. A huge
government project aims to link every home and business by fibre-optic cable
by 2015.

Communications stand to gain most because optical transmission is quicker
and more reliable than electrical transmission. ‘In the future, silicon
devices will have to be assisted by optical interconnection,’ says Izuo
Hayashi, who has just retired as director of the gov-ernment’s Optoelectronics
Technology Research Laboratory (OTL) after 25 years working in the field.
Before that, he worked on radiation detectors; then Bell Telephone Laboratories
invented the transistor. ‘I grew up in the age of vacuum tubes, where you
could see things coming off the cathode,’ Hayashi recalls. ‘But the transistor
was tiny, tiny. I was intrigued by what was in that small crystal. I was
young . . . oh, about 40.’

Since then, the crystals have shrunk dramatically, but he has kept his
grasp on the subject. ‘Growing gallium arsenide (the basis of semiconductor
lasers) on silicon substrates will be important, but nobody has managed
it because the crystal lattices are different . . . though lots around the
world are trying. I don’t believe you’ll ever replace silicon technology
– it has been and is ideal, for switching and amplification, and optoelectronics
can’t compete with that.

‘But on the other hand, electronic circuits are weak for transferring
information even over 1 centimetre. Optoelectronics is better there.’ Because
light beams do not need to be insulated from each other, replacing electrical
wires with optical ones would make it easier to shrink processors and computers.
Given two printed circuit boards covered in chips which have to communicate
with each other, it is a lot easier – in theory – to make complex connections
using laser light rather than wires which have to be soldered, bringing
the risk of high temperatures damaging delicate components during manufacture.

Systems designers have also sought for a long time to minimise wiring
in conventional printed circuits by reconfiguring the flat boards as cubes.
This three-dimensional approach could speed systems up, but there are two
problems: the wiring quickly becomes extremely complicated and dense, and
this, allied to the reduced open surface area, leads to overheating. Hayashi
thinks that optoelectronics could solve both problems: ‘An elementary calculation
shows that replacing metal wires with light makes it possible to change
the circuit design, and operate processors in parallel.’

OTL’s mission – to develop future technologies in optoelectronics –
is a long-term one; Hayashi cheerfully looks to the next century before
anything comes of the projects. Growing gallium arsenide crystals on a silicon
base is one of the main objectives, which could eventually lead to micrometre-sized
lasers and optical detectors. Other projects include using finely focused
electron beams to grow crystals in the shape of quantum wires and even quantum
‘boxes’, two-dimensional quantum wires able to hold one electron at a time,
like a memory storage device. And it is memory devices that the processors
of the future will need to seek and send data.

At the moment, the currency in the electronics world is memory chips
able to store 4 megabits. Chips able to store 16 megabits are ready to hit
the market, and today’s chip designers have their sights set on 256-megabit
chips and, for the next century, on 1-gigabit and 4-gigabit chips. ‘The
consensus is that 1-gigabit chips will be in use around 2005 or 2010,’ says
the general manager of NEC’s fundamental research laboratories, Roy Lang
(or Hiroyoshi Rangu, as he also calls himself, depending on whether he is
stressing his New Zealand or Japanese ancestry). ‘Experts have different
opinions about whether the 4-gigabit chip is actually feasible. It needs
more perfect silicon crystals (on which to etch the transistors forming
the memory elements). But looking back at history, I think people will overcome
the problems.’

Sakaki is less sure. Elements as small as Lang is proposing run into
quantum effects: the electron has a small but finite probability of ‘tunnelling’
through solid matter and appearing, unwanted, on the other side of a transistor
gate. ‘The quantum phenomenon has to be understood and controlled,’ says
Sakaki. ‘Up to three or four years ago, silicon engineers didn’t have to
worry about it. But the designers of the 256 megabit chips are fighting
it. Tunnelling is becoming an important phenomenon, and problem, in large-scale
integrated circuits.’

Controlling the atoms

Perhaps the answer is not to use electronic methods at all, but atomic
ones. A major proponent here, and equally eminent in the field of nanotechnology,
is Masakazu Aono. He is head of the Atomcraft project, which aims to find
ways to control the structure of solid surfaces down to the atomic level
and is funded by the government’s Science and Technology Agency.

Like Sakaki, the eminence of his position is not reflected by the grandeur
of his office, which is also stuffed to capacity, and in which everything
except the occupant looks slightly the worse for wear.

Aono’s work has been transformed by the invention of the scanning tunnelling
microscope, or STM, by IBM researchers in 1981 which can deposit or pick
up individual atoms on a substrate by moving its tungsten tip around the
surface. ‘In theory, we can store all human knowledge on a single chip by
using a single atom to represent one bit,’ says Aono. Which sounds fine,
until you consider that – assuming it takes one second to write one bit
– storing the works of Shakespeare (about 8 megabytes) would take the hapless
operator more than two years, not including sleep. Aono recognises that:
‘You don’t have to use just one tip – you could have an array, so rather
than taking a hundred years to record sixty years’ worth of information,
it would only take one year.’

The only problem is that an STM is not so easily shrunk. By its nature
it requires accurate positioning and sophisticated machinery which right
now would not fit happily in a bus, never mind in the pocket beside the
hardback-sized supercomputer you got for Christmas 2010. But that is not,
for the moment, a deterrent. As Aono points out, things have already changed
rapidly. ‘Ten years ago we didn’t have STMs, so it was impossible to manipulate
single atoms.’ Reading the first paper about STMs, he was inspired. ‘I saw
immediately that this was the way of getting closer to surfaces. Three years
after reading that paper, we started building our STM.’

They had some uncomfortable times on the way there. Oddly enough, among
nano-scientists, the tussle for supremacy is fought, like a teenage gang
skirmish, via graffiti. Don Eigler, a researcher at IBM’s Almaden Research
Center in San Jose, California, used an STM in 1990 to write the company
name with 35 atoms of xenon. Shojiro Asai at Hitachi’s central research
laboratory replied in 1991 by etching a much miniaturised version of the
four lines within a circle that form the symbol for Peace. For a while,
Aono lived in fear of the competition. ‘It took the people at IBM maybe
seven or eight years to invent the STM – we had no knowledge about it, so
we took a while over building our own,’ he recalls. ‘After that, I organised
the writing of a paper about our results for Atomcraft . . . which is when
I got frightened. For a while, I was almost scared to pick up Nature and
Science in case of what I might read – that someone had beaten us.’

But Aono was soon up among the elite with a string of his own publications
in those journals. Sakaki is not convinced, however, that Aono’s proposed
method for building storage devices with atoms or making quantum boxes by
STM is entirely practical. ‘I think that’s an extreme technology, manipulating
atom by atom,’ he says. ‘Making even one quantum box means manipulating
125 000 atoms, and for a storage device you need hundreds of them, which
means controlling about 10 million atoms. Whether the STM is practical
. . .’ He pauses. ‘I’m not critical, because in this field every approach
has to be tried.’ But it’s clear that Sakaki is not volunteering to be in
the first shift making quantum boxes by STM, either.

There is another, entirely different approach, to high-density memory
devices now being devised by Canon. It is probably the highest-density storage
method to be within striking distance of full-scale production.

Canon occupies an odd position among Japanese companies. According to
Takashi Nakagiri, who runs Canon’s central research centre, researchers
are under standing instructions to follow developments that are emphatically
not part of mainstream thinking. In the 1970s when the world thought that
IBM’s patented droplet printing-head method was the best for computer printers,
Canon’s scientists wondered about other techniques. The result was the bubblejet
printer – quieter, cheaper and more effective.

So now, while everyone else is puzzling over how to make transistors
a few angstroms across, Canon is working on a memory device based on Langmuir-Blodgett
films, in which the molecules consist of a hydrophilic head (which is attracted
to water molecules) and a hydrophobic tail (which repels water). In washing-up
liquid, such molecules make water and fat miscible: they form a layer one
molecule thick at the water-fat barrier as the hydrophilic ends attach to
water molecules while the hydrophobic ends attach to fat molecules.

At Canon, where a team has been working on such films for ten years,
they are used to form a layer one molecule thick between two other insulating
substrates. One substrate starts off wet, so the molecules stand vertically
above it with their hydrophilic heads on the surface and the hydrophobic
tails in the air. The other substrate is then placed on top. Canon has a
prototype with memory locations about 25 nanometres (about 100 molecules)
apart – the equivalent of a memory density of 1 terabit (10 ¹²
bits) per square centimetre.

The device plays on a peculiar property of the films, discovered by
Canon’s Kiyoshi Takimoto in 1985, which is that they show bistable conductance
(see Figure 2). If a voltage is applied through an STM point, no current
flows at first. Then, suddenly, the resistance of the molecules falls and
current flows – and can keep flowing even if the voltage is reduced. The
point’s conductance has altered, and remains different from its surroundings
even if the voltage is turned off. If a higher voltage is applied, the conductance
breaks down and returns to the original state. This turns each point into
an erasable memory, storing one bit, and the location of the bits shows
clearly in an STM conductance ‘map’.

Up to speed

The advantage of this method over those proposed by Aono or Sakaki is
that Canon can already coat substrates with the film at production speeds
– 10 square metres an hour (that is, 10 ¹⁷ bits, or a hundred
million gigabits). Atomic excavation is not required; there are no tunnelling
electrons to worry about. Of course, an STM is needed to read, write and
erase the data, but the principle is clear.

Japan’s engineers thus have at least the ideas for the interconnection
and storage devices of the computers of the future. But what will they be
able to do?

Japan has some pride at stake here. Its much touted Fifth Generation
project was a ten-year programme to make the country a world leader in artificial
intelligence software which would run on super-powerful Japanese parallel-process-ing
computers. In fact it achieved few of its targets and fizzled out, a 54
billion yen flop.

But the Japanese are nothing if not dogged. The Ministry of International
Trade and Industry has now launched another 10-year marathon, the Real World
Computing Project, which will cost 70 million yen. Junichi Shimada, a veteran
of such work, is its director. An avuncular chainsmoker with a wry view
of technologies, he has worked at MITI’s Electrotechnical Laboratory, set
up the OTL a decade ago, and jokes: ‘My speciality is as an architect of
laboratories.’ In stark contrast to Sakaki or Aono, he has a plush office
in the heart of Tokyo.

The aims of the RWCP are blurred at best. ‘We’re aiming at a kind of
flexible information processing,’ says Shimada, lighting another cigarette.
‘We call it ‘soft logic’.’ He cites the cocktail party effect in which humans
can concentrate on one quiet conversation in a noisy room. ‘In the physical
sense, the information level is below the noise level. You need soft logic
to extract the information, inferring what may not be obvious.

‘Another important aspect is that a real-world computer must be real
time. In an exam, when your time is up, you have to submit your paper. It
might not be perfect, but it’s too late. Computers never want to give in
an imperfect paper. But being real time is always associated with incompleteness.’

Perhaps because having concrete targets caused trouble for the Fifth
Generation project, MITI is taking a more careful line this time. ‘We hope
in five years’ time we can set a goal. But doing that is 80 per cent of
the work, and we want it to be comprehensive. Industry people talk about
nothing but goals, but you notice that until one is set, they don’t make
a move, especially in Japan.’

There are 20 industrial members in the consortium, including electronics
and computer companies such as Hitachi and Fujitsu. There are also members
from Germany, Sweden, the Netherlands and Singapore, following pressure
on Japan to open up its markets and research.

Lessons have been learnt from the Fifth Generation: where more than
half its cash went on hardware, the RWCP will spend 90 per cent on software.
Shimada grins. ‘Pattern processing may be something we can do well. Kanji
(the Chinese characters used to write Japanese) is much more a pattern than
an alphabet, and the structure of our language isn’t logically oriented
– it’s a fuzzy language. So the way we communicate is fuzzy compared to
the Western style.’

Fuzzy language, fuzzy aims. It falls to Sakaki to see what Shimada and
the others are trying to do in a wider context. ‘Humans always try to control
different types of materials,’ Sakaki notes. ‘Starting from huge stones
in the dawn of time. This is just the same, only on a smaller scale. What
we want to do is push the technology to its limits. Even when we fail, we
get useful data. In some ways, we’re just like the European explorers going
into Africa.’