91av

HAS silicon had its chips? The next generation of computer processors could
be grown in vats of chemicals and be millions of times faster. With components
forged from single molecules, even the tiniest chips could be vastly powerful.
Already chemists have succeeded in making individual components. Now they’re
working out how to stick them together. And it’s stretching their ingenuity to
the limit.

So what’s wrong with silicon? Well, not a lot. But if miniaturisation
continues at its present pace, by 2008 silicon devices will probably be as small
as they can get. Shrink things too much and electrons will jump from one
component to another—the circuits simply won’t be able to contain their
electrons.

Overcoming this barrier to continued miniaturisation is causing a lot of
head-scratching among scientists. Some think the best hope is to abandon digital
computers altogether and invest in more radical designs using DNA or bizarre
quantum effects. But don’t dismiss digital computers just yet. A growing number
of scientists believe there are ways to shrink computer components by an
astonishing million-fold. And to do it they’re using some of the smallest
building blocks around—molecules.

“If it works, there are two important advantages,” says James Ellenbogen of
the MITRE Nanosystems Group in McLean, Virginia. “It makes use of a lot of good
stuff we already know about building and programming digital computers. And it
allows us to go to the very limits of miniaturisation.”

To get an idea of just how small that could be, consider the differences
between today’s micro-devices, such as transistors, and molecules. A modern
computer contains about a billion transistors. If everyone on the planet had 10
computers, they would contain 60 billion billion transistors in total. But there
are ten thousand times as many molecules in a single gulp of water. Of course,
water molecules aren’t computer components, but researchers have made other
molecules that are. And they don’t think it’ll take long to build them into
molecular computers.

The idea of forging electronic components from single molecules was first
proposed in 1974 by Ari Aviram of IBM’s T. J. Watson Research Center in Yorktown
Heights, New York, and Mark Ratner, now at Northwestern University in Illinois.
But it took researchers until 1996 to make a key breakthrough. By adding
sulphur-containing “thiol” molecules to the most basic of molecular
devices—a wire—they were able to anchor wires to a gold sheet and
for the first time, run currents through them.

These wires, designed by Jim Tour of Rice University in Houston, Texas, and
his colleagues at Pennsylvania State University were based on a molecule called
polyphenylene, essentially a long chain of carbon rings. In most molecules, the
electrons that bind them together move in well-defined orbitals. But in
polyphenylene, these electron orbitals overlap, forming a conducting channel
along the length of the molecule (see Diagram).
To get a current to run down one of these “Tour wires”, you just have to apply a voltage across it
(91av, 2 August 1997, p 32).

A wire is fine, but by itself it isn’t going to get any computing done. For
that you need switches. So Tour and his colleagues set about making one. In
1997, they came up with a molecular version of a diode—a type of switch
that works by letting current flow in one direction only. To make it, Tour took
two polyphenylene wires and linked them together with an insulating molecule,
such as methylene. To one wire he added two “methoxy” molecules that pushed
electrons into the wire. To the other, he added two “cyano” molecules that drew
electrons out of the wire. This makes current flow far more easily in one
direction than the other. By linking these together to make a molecular diode,
you can make some useful electronics.

That’s exactly what Ellenbogen and his colleagues are doing back at MITRE.
Using molecular diodes, they’re designing logic gates—the electronic
circuitry needed to crunch numbers
(see Diagram). As well as common logic
gates such as AND, NOT and XOR, they’ve drawn up plans for a complex molecular
“half adder”, a fundamental computer circuit that takes two binary digits and
adds them together. On paper, the half adder may look like a sprawling
super-molecule, but it takes up only 100 square nanometres—at least a
million times smaller than its silicon counterpart.FIG-mg22374301.JPG

Of course, no computer would be complete without a memory, but this is an
easier problem to solve—you just need a bank of on/off switches. Last year
Tour and Mark Reed at Yale University revealed just such a device—a simple
molecular memory that could be written to and read electronically
(91av, 13 November 1999, p 23).

Throw all these components together and you’ve got the essentials for a
working computer. So where can you buy one? Well, nowhere right now, because a
lot of thorny problems still need to be overcome, like how to actually make
one.

Scientists know that the molecules they’ve created should work as electronic
components. But that’s not the same as building them. Armed with a good
knowledge of organic chemistry, however, and there’s no reason why researchers
shouldn’t be able to produce millions of these molecules. The next stumbling
block will be connecting them all together.

One way to do this could be to use a machine called a scanning tunnelling
microscope. This has a very fine tip that can push molecules, or even atoms,
around—as demonstrated in 1990 when IBM researchers spelt out their
company name with 35 xenon atoms. But there are two problems with direct
manipulation. Some molecules stick to the surface they’re on and break apart.
Others simply go flying when you touch them.

Even if molecules could be pushed into the right positions one by one, how
long would it take to hook up the 10 million needed to make the molecular
equivalent of a Pentium chip? There might be a way to automate the process, but
it would take some doing.

A more complex, but far quicker and more elegant approach could be to make
the molecules “self-assemble”—arrange themselves into the circuits you’re
trying to build. With thiols attached like crocodile clips, Tour made his wires
stand up on gold sheets. But making complicated circuits with different
components is a much bigger problem. “The more complexity you want out of them,
the tougher it will get,” says Tour. But there’s worse to come.

Not fade away

You’ve still got the problem of how to achieve “gain” in a molecular
computer. Electrons lose energy as they zip around circuits, so currents need to
be boosted to make sure they don’t become too weak. In electronics, transistors
hooked up to a separate power supply top up weak currents, but so far, no one
has created a molecular transistor. Without that, the already faint signals in a
molecular circuit could easily drop below a useful level.

Another difficulty is a result of miniaturisation itself. How would you
connect a keyboard and screen to such a tiny processor? In conventional
computers, this is fairly straightforward. But when you’re trying to connect
things to the nanoworld, where components and currents are orders of magnitude
smaller, it won’t be so easy. Connecting individual molecules to each other will
be tricky too.

Once you start trying to get billions of devices hooked up together, the
number of interconnections shoots up, and with it the potential problems. Since
these devices will be so small, tiny defects could cause big problems. Even if a
single molecular device in a billion has a dodgy connection, a chip with a
billion devices would throw up errors. But designers may be able to work around
this problem.

According to Terry Fountain, an expert in computer architecture at University
College London, molecular computers must be built to cope with defects such as
faulty circuitry and defective components. Each component, he says, will need as
many as a hundred back-up copies—built-in “redundancy”—to ensure
that at least one copy works. But this raises an obvious point. “If you have to
build a hundred transistors to get one to work, maybe you shouldn’t build them
so small in the first place,” he suggests.

Others are more hopeful. A collaboration between James Heath’s group at the
University of California, Los Angeles, and Phil Kuekes’s team at
Hewlett-Packard’s labs in Palo Alto, California, resulted in a computer that can
tolerate defective components
(91av, 7 November 1998, p 50).
The Teramac, as it is known, was built with standard silicon devices. The
difference is that Heath and Kuekes deliberately used thousands of faulty
components, but still ended up with a working computer. Teramac’s circuits have
a special layout that allows the computer to redesign itself. First it tests
which of its circuits work and which don’t, then it finds a way to route signals
around the damaged parts.

A similar design might work for self-assembled molecular computers. “The
important lesson of Teramac for nanotechnology is that a system doesn’t have to
be perfect to be very powerful,” says Stan Williams, head of basic research at
Hewlett-Packard. “And the more defects a system can tolerate, the cheaper it
will be to build.”

A simpler way to go, according to Tour and Keukes, might be to give up trying
to create chips according to specific designs. Instead, key components and their
connections could be “grown” randomly using self-assembly. Each chip would be
different and would have to be tested to see what inputs produce what outputs.
Some would probably not work at all, while some would work better for certain
applications than others. “You could say, `Hey, this one’s great—let me
use it in a Cray computer.’ The next one, say there are a lot of faulty
circuits, you would put that one in a toaster,” says Tour.

Ironically, even if scientists work out how to make molecular components,
they might not be able to take full advantage of the miniaturisation they
promise. The problem is heat. With all those electrons charging around inside
them, even the relatively gigantic chips in your personal computer heat up.
Molecular circuits packed a million times more densely will get extremely
hot—a processor could get as hot as a stove. “It will fry,” says Tour.

One hope is that some yet-to-be-discovered material capable of removing
excess heat could prevent molecular computers from disintegrating into
smouldering blobs. Tour, however, proposes another strategy. He believes
molecular computers could work without a single electron having to move around a
circuit.

Feel the force

Instead of using electric currents, Tour says signals could be passed from
one component to another using electrostatic potentials. “Suppose you have an
electron in your left hand and an electron in your right hand,” says Tour. “As
soon as you bring them close together, they’ll feel a repulsion. Likewise, if
you bring a charge up next to a molecule, you’ll cause a shift in its electric
field.” The change in the electric field could be used as the signal. And no
moving electrons means no heat.

The idea isn’t new, Tour points out. “This is how molecules talk to each
other in nature—they sense changes in each other’s electrostatic
potential.” Tour has good reason to believe the same principle could work in
molecular computers.

A group led by Craig Lent at the University of Notre Dame in Indiana has
already proved the principle. “He’s built an electrostatic working system,” says
Tour. “It’s very large—micron sized. What we’re saying is we can
extrapolate that down to the molecular scale.” And doing so wouldn’t require
brand new materials. “All molecules have the potential to work
electrostatically,” says Tour. “Some are just better than others.”

As well as getting around the problem of overheating, electrostatic computing
has other benefits. “This sort of interaction is around six orders of magnitude
faster than where we are now with a typical current in a computer,” says Tour.
“It will literally be lightning fast.”

Other strategies for preventing overheating are being investigated. One is to
pack molecular circuits less densely. This would appear to defeat the object of
molecular computing, but it’s not as absurd as it sounds. You could leave plenty
of room between molecular components and still pack them one hundred thousand
times more densely than today’s chips.

Perhaps even curiouser is Ellenbogen’s suggestion to run chips slower than
ever. He says that by packing molecular circuits very tightly and designing them
for specialised tasks, you could run each device very slowly and still get huge
computing power. In the brain, neurons fire a modest 100 times a second, passing
signals 5 million times slower than a modern computer. But since neurons are
jammed together so closely, and brain structures are so specialised, you end up
with an extremely effective computer. The brain, though, is adept at processing
many signals at once—parallel processing—which accounts for much of
its power.

With all of the technological advances still needed to make molecular
computers work, you could be forgiven for thinking it’s a hopeless task. But
Ellenbogen disagrees, pointing to the history of silicon chips. “Etching a
circuit in a silicon chip is complicated. It requires very sophisticated
chemistry,” he says. “The only reason it looks simple to us now is because we’ve
been doing it for 40 years.”

Tour believes the remaining hurdles are surmountable, and even predicts that
hybrid systems using traditional and molecular components will appear within two
years. “What makes me think it’s going to work? All the pieces point to it
working,” he says. “We can self-assemble these things—order them on
surfaces—we can make working logic gates out of them. We are doing this,
we are making this work. It’s not a matter of what makes me think this’ll work.
It’s working already.”