Silicon has become synonymous with microelectronics, but now any
number of pretenders are waiting to take its place, says Joerg
Heber
by Joerg Heber
IN 1965, a year before the first pocket calculator was invented, a
young physicist from Silicon Valley, Gordon Moore, made a daring
prediction. He claimed that the number of components squeezed onto a
single silicon chip would double about every two years. And double,
and double and continue to double. If he had been right, the best
silicon chips today would contain an unbelievable 100 million single
components.
The true figure is more like 2 billion: Moore had underestimated how
fast the shrinking trend would take off. Since the mid-1970s, though,
his "law" has been a bankable certainty, influencing economic, social
and scientific developments in ways that are hard to overstate.
Google, genome sequencing, multiplayer video games, the search for the
"God particle": all rely on silicon's seemingly limitless ability to
deliver more computing speed and capacity - for an ever-diminishing
price.
Moore's law means that the cellphone or iPod in your pocket today has
more byte-crunching power than the mainframes on the Apollo
spacecraft, at the planning stage when the prediction was made. Had
aviation followed a similar trajectory, a flight from New York to
Paris in 2005 would have cost a cent and lasted less than a second,
Intel - the chip giant Moore co-founded in 1968 - calculated.
Can the trend go on? Reports of the imminent death of Moore's law have
been around almost as long as the law itself, and have always proved
exaggerated. But now there is concrete cause for concern. The smallest
features on today's state-of-the-art chips are just a few nanometres
across. At the current rate of shrinking, they will reach the size of
a few silicon atoms by about 2020.
At this kind of scale, the properties that make silicon the
microelectronic material of choice will fail. We must then either
abandon silicon and find an alternative, or accept that the
ever-increasing computing power we have come to depend on has reached
its upper limit - until the unforeseeable time when more exotic
computing schemes, such as quantum computing, become commercially
viable.
And so, even as the last drops of computing power are squeezed out of
silicon, the race is on to find its successor. Bit by bit, a startling
picture is emerging. Not only might natural processes hold the key to
the computer's further evolution, but the ideal nanoelectronic
material might also have been under our noses all along. It could well
be nature's own favoured building block - carbon.
Silicon squeeze
It is easy to forget that no natural law says chips equal silicon. The
element wasn't even the first-choice microelectronic material. It is
relatively difficult to obtain in the very pure form needed, and was
initially passed over in favour of its neighbour in the periodic
table, germanium. In the end, though, silicon's reliability,
performance and abundance - the basic material is freely available in
sand from most beaches - won out.
Germanium and silicon are neither insulators that refuse to conduct
electricity at all, nor conductors that let electrons flow all too
freely. They are semiconductors: reluctant to conduct in their natural
state, but requiring just a nudge, in the form of a small voltage
applied to them, to be persuaded. That sweat-free swapping between
conducting and non-conducting states makes semiconductors the ideal
materials for transistors, the basic on-off switches that make up
electronic logic circuits.
The first transistors were rather crude and bulky devices, with large
wire contacts pressed against an underlying slab of germanium. Today,
lasers are used to etch them in their billions onto "wafers" of
ultra-pure silicon just a few micrometres thick. Over the past 20
years, Robert Chau has seen efforts to squeeze this compact construct
get ever more challenging. He leads research into advanced transistor
designs at Intel, which supplies about 80 per cent of the world's
silicon processors. "Until the 1990s, Moore's law was continued by
relatively simple scaling," he says. Now his researchers are having to
change the materials: almost every part of the transistor that was
originally silicon is being replaced by something else.
Take the two contacts through which electrons flow into and out of the
transistor, the source and the drain. Traditionally, miniaturising
these was a question of finding lasers with smaller and smaller
wavelengths to etch out ever tinier features on the chip. In 2003,
Chau and his Intel colleagues started to use an alloy of silicon
and germanium for the contacts. The separation of atoms in the alloy
is larger than in pure silicon, and the atomic bonds in the channel
become slightly stretched to span the extra space. Electrons can nip
more speedily through this strained silicon, allowing more information
to be processed by a smaller number of transistors.
Or take the gate. This is the transistor's on-off switch, which sits
atop the channel separated by a thin insulating film of silicon
dioxide. This film must be as thin as possible so as not to slow down
the transistor's response when a switching voltage is applied to the
gate. But at just a few atoms thick, it is now about as thin as it can
go. The route to further miniaturisation is again to break silicon's
monopoly on the chip, adding different, sometimes exotic, elements
into the transistor brew. The very latest chips use a superior
insulator, hafnium dioxide, for their gates.
Then there is the general transistor layout. Traditionally, this is a
low-rise sprawl that takes up valuable real estate on the wafer. The
obvious solution is to do what we do in city centres - build upwards.
Intel has made prototypes of what it calls trigate transistors, in
which the gate, rather than lying on top of a flat channel, is wrapped
around three sides of a raised channel. The larger contact area makes
the gate more effective, increasing the switching speed and allowing
the same number of transistors to do more work.
Chau thinks the trigate is such a breakthrough that it will become a
permanent feature of any future transistor. Put it all together, and
he is confident that enough life can be squeezed out of silicon for
the next two or three generations of Moore's law. "Beyond that, our
view gets more fuzzy," he says.
Wei Lu, an electrical engineer from the University of Michigan in Ann
Arbor, thinks we can see a little further - provided we turn
established ways of thinking on their head. Currently we take a
"top-down" approach to microelectronics, taking a silicon wafer as the
raw material, then carving ever smaller components into it. But the
burgeoning field of nanotechnology allows us to experiment with the
kind of thing nature does all the time: building things from the
bottom up by letting atoms self-assemble into tiny structures.
"This approach significantly broadens the choice of materials that can
be used," says Lu. Atoms of almost any element, including silicon, can
be coerced into forming nanometre-scale structures. In effect, you
just throw the required ingredients - molecules with the right shapes
or with particularly desirable electronic properties - into the mixing
bowl, and let chemistry take its course.
Back to basics
There is a catch. Nature does not have the same need for quality
control as the electrical engineer: it tends to build as it pleases,
and if one result is slightly different from the next, that's all part
of the fun. The uniformity needed for transistors is a lot more tricky
to achieve when building them up from scratch than it is when using a
template to etch out identical transistors on a silicon wafer.
Tricky, but not impossible. Lu champions a technology called the
crossbar array, which sidesteps the replication problems. One of the
most useful types of array, pioneered by Charles Lieber and Yi Cui of
Harvard University, is made up of a series of self-assembled
semiconducting nanowires laid over each other at right angles to
create a square mesh. In this arrangement, whether individual wires
are exactly the same ceases to matter (Science, vol 291, p 851).
What matters is whether two wires at a junction have the same voltage
signal across them. If they do, the junction conducts, and assumes a
logical "on" state. Conversely, if the two signals are of different
strengths, or if there is no voltage on either, that particular
junction is switched off. The junction thus reproduces the basic
controlled switching function of a transistor. An array of them can be
wired up, just as a group of transistors can, to produce all the main
types of logic gate needed to make an integrated circuit.
A promising brand of crossbar array uses molecules known as
rotaxanes for its switching junctions. These consist of two
components, looking rather like a dumb-bell with a ring encircling its
midriff. The relative positions of these two molecules, easily changed
by a small voltage, determine whether the arrangement conducts or not.
Rotaxanes have been used to create some of the densest crossbar arrays
yet, cramming in junctions as densely as the electronics industry
would like to fit transistors on a chip by 2020 (Nature, vol 445, p
414).
The bottom-up approach to building microelectronics has collateral
benefits: it can produce materials that do things silicon never could.
Semiconducting polymers built from organic molecules can be used to
make transparent, flexible electronic circuits that would be
impossible with brittle, opaque silicon. If that sounds frivolous,
just consider the possibility of electronic newspapers that you can
roll up and put in your pocket, or solar cells that fit on your
curtains.
Signal speed could also benefit from the approach. Currently, a lot of
computing time is wasted transporting electrical signals between
different parts of the chip. What if that transport role could be
taken over by light? Normal silicon does not emit light, but nanoscale
particles of the element can be made to. The incorporation of
nanoscale silicon light emitters and absorbers would provide the same
kind of leap in speed that optical fibres give over copper cables in
an internet connection, and keep Moore's law going independently of
the size of transistors.
All such advances are likely to be integrated into chips in the coming
years. But they are stopgaps: they might maintain Moore's law for a
while, but sooner or later, the limitations of silicon will make a
wholesale switch to a different base material unavoidable.
What material that will be is less certain. Andre Geim, a physicist
from the University of Manchester, UK, has seen successors come and
go. "Eventually, they all failed miserably," he says. Now, though, he
thinks he might have the answer in a material he and his colleagues
discovered in 2004. If his hunch is correct, the future of computing
could lie in shavings of carbon just one atom thick, a material known
as graphene.
Other scientists are also coming round to the idea. Lu is one. "It's a
definite contender," he says. Firstly, its base material is easy to
get hold of. Make a pencil sketch on a piece of paper and - if you
happen to have an electron microscope to hand - here and there you
will see a single-layer sheet of graphite among your scribblings. That
is graphene. The microscope will reveal a two-dimensional honeycomb
structure of six-atom carbon rings.
What it will not reveal is the material's quite astounding properties.
Electrons can travel along its single layer, like bullets from a gun,
without hitting the obstacles they would encounter in a 3D structure.
A graphene sheet is mechanically extremely stable, so a transistor
channel could be just a few interlocking carbon rings carved from it.
Graphene's atomic bonds, like those of that other notoriously hardcore
form of carbon, diamond, are also supremely strong, allowing very
many electrons to flow across them simultaneously - essential if you
want to send useful electric currents through a transistor channel
consisting of just a few atoms.
Normally, graphene sheets are metallic, making them far too easy on
electron movement to allow the kind of subtle switching that
transistors depend on. To encourage the necessary semiconducting
behaviour, the electrons must be reined in a little by cutting the
graphene sheet into pieces. Large pieces of graphene can only be made
semiconducting by lowering their temperature significantly, but the
smaller the piece the higher the temperature at which the switch-over
happens. At room temperature, the critical scale is about 10
nanometres.
Coincidentally, that is just about the scale at which silicon
technology becomes unstable. Once the limits of silicon
miniaturisation have been reached, graphene could therefore be ideally
poised to take on the Moore's law mantle. Geim and his colleagues have
already shown the potential of the technology, creating the smallest
graphene transistor yet earlier this year, with contacts made of
metallic graphene and a channel of semiconducting graphene just 30
nanometres across (see diagram). They also achieved room-temperature
operation - albeit somewhat erratically - with even smaller pieces
(Science, vol 320, p 356).
The challenge is clear if carbon is to become the new silicon:
mass-production fabrication technologies are needed so that graphene
transistors can be reliably and reproducibly made. "We have a material
which allows you to go to molecular-scale electronics," says Geim.
"But we don't yet have the tools."
We already have a material which allows us to go to molecular-scale
electronics
Change will come, but silicon is so entrenched that the price for
abandoning it is huge. For that reason, Lu thinks that the coming
years will be more evolution than revolution. "I expect we'll see some
hybrid circuits first, with new materials doing things such as
high-performance logic," he says. In those hybrid chips, silicon will
gradually be reduced to a peripheral role, performing housekeeping
functions such as controlling data flow and timekeeping.
The urgency of efforts to improve and replace silicon underscore our
insatiable demand for computing power. Moore's law has become a
self-fulfilling prophecy as chip companies invest huge sums to ensure
its promise is maintained. Soon, it will enter a new phase, decoupled
from the material with which it has been synonymous for 40 years. No
longer a measure of the development of silicon alone, Moore's law will
become a yardstick of our progress as we harness the cunning of
nature's design strategies.
Joerg Heber is an editor of Nature Materials
To unsubscribe send a message to [email protected] with
the subject unsubscribe.
To change your subscription to digest mode or make any other changes, please
visit the list home page at
http://accessindia.org.in/mailman/listinfo/accessindia_accessindia.org.in
