杏吧原创

鈥淔ASTER, faster,鈥� goes the eternal cry of the impatient computer user. In a
world dominated by the processor, speed is everything. To feed our addiction for
powerful computers, design engineers are resorting to all manner of exotic
solutions: manipulating the quirks of quantum mechanics, sending data along
twinkling beams of laser light, or even using chips cooled to near absolute
zero. But the latest idea for opening up the data throttle uses a trick that
nobody has considered till now: doing away with currents of electrons inside
computer chips, and making waves instead.

At Rensselaer Polytechnic Institute (RPI) in New York, the physicist Michael
Shur is leading a team of researchers investigating this idea. Their aim is to
build tiny components that will handle electrical signals with frequencies in
the terahertz (10¹² hertz) range. These are far higher frequencies than are
possible with conventional microelectronic components鈥攕o high, in fact,
that they are encroaching on the area that belongs to infrared optics. What鈥檚
more, these devices will do more than process data at super-high speeds. They
could also act as sensitive detectors able to spot minute quantities of drugs,
poisons, pollutants or explosives. They could even form the basis of a new type
of camera that could peer beneath clothing and see through walls.

As the researchers readily admit, they are setting foot in unknown territory,
where the theories and design tools that work for conventional semiconductor
devices start to break down. 鈥淲e are working at the intersection of two very
challenging fields,鈥� says Shur. 鈥淥ne is making submicron devices, and the other
is working in the terahertz frequency range, which is new to most people.鈥�

At heart, a computer chip is nothing more than a complicated arrangement of
transistors, which act as electronic on/off switches. The ultimate speed of a
chip is limited by the rate at which you can flick those switches. This in turn
depends on the time it takes for individual electrons to cross from one side of
the device to the other鈥攊n contrast to current flowing through an ordinary
conductor, where the signal travels much faster than individual electrons
drift.

What Shur and his colleagues are investigating is a way to break through this
speed barrier. They reason that as transistors become ever smaller, the
electrons slopping around inside form a fluid-like plasma. And in a plasma you
can generate waves that move far faster than a dawdling current of electrons
ever can. They hope that minute transistors rippling with plasma waves could
boost computer processing speeds up to 100-fold.

Some 95 per cent of the transistors in today鈥檚 computers use field-effect
transistors, which rely on an electric field to flip between their on and off
positions. Each FET has two contacts鈥攖he source and the drain鈥攚hich
connect to regions of electron-rich semiconductor material. They are separated
by a channel made from an electron-deficient semiconductor
(see Diagram).
A third contact, called the gate, runs along the top of the device but is
isolated from the channel by a thin insulating layer. Putting a small charge,
known as the bias, onto the gate contact generates an electric field which
switches on the current flow through the transistor. 鈥淲hen we apply a positive
voltage to the gate, electrons rush into the channel and connect the source and
the drain. This is the transistor鈥檚 `on鈥� state,鈥� Shur explains. 鈥淲hen the gate
voltage is more negative or zero, we have no electrons in the channel. This is
the `off鈥� state.鈥� These two states constitute the ones and zeros that are
fundamental to digital computing. The speed at which a transistor can flip
between the two states will limit the total number of calculations that the
processor can handle each second.

Until now, electronics engineers striving to speed up transistors have gone
for the obvious solution: they shrink the device, so the sluggish electrons
don鈥檛 have so far to travel. Today鈥檚 transistors are typically 0.25 micrometres
long鈥攁bout 400 of them would fit across a human hair鈥攁nd future
devices may shrink to a minuscule 0.1 micrometres or less. But Shur and his team
suspect that this will bring only limited performance improvements. 鈥淭ransistors
will reach 0.07 micrometres or so within the next decade, with diminishing
returns for all this technological prowess,鈥� says James Lu, who works with Shur
at RPI. Fundamental problems such as quantum noise will limit this approach, he
says. There are other difficulties too: just making the contacts small enough to
link these tiny transistors together is a major challenge. Before long, a
radically different approach will be called for.

Semiconductor physicists have become used to describing the electrons in
today鈥檚 transistors as akin to a gas: they behave like isolated particles,
drifting through the device, occasionally bumping into each other. But Shur has
calculated that as transistors shrink further, the number of electrons in a
given volume of the material will rise to a point where they start to behave in
a completely different manner. No longer will they act as a gas, Shur says.
Instead, he believes, they will resemble a fluid, and it is the behaviour of
this fluid that is the key to plasma-wave formation.

To their surprise, Shur and his colleague Michel Dyakonov of the Ioffe
Institute for semiconductor physics in St Petersburg, Russia, have found that
this electron fluid behaves exactly like water trapped in a shallow channel.
Push the water at one end and longitudinal waves鈥攁lternating compressions
and expansions鈥攔ush along the channel faster than an individual water
molecule could.

Using plasma waves to transmit signals offers a similar advantage: the plasma
waves move much faster than a single electron can. 鈥淓lectrons drift through a
transistor at about 10⁵ metres per second, but for a plasma wave, we expect
maybe 10⁶ metres per second or higher,鈥� says Shur. Devices relying on drifting
electrons could operate at up to 400 gigahertz, Shur estimates, but plasma-wave
devices could run more than 20 times faster鈥攁t frequencies approaching 10
terahertz.

Shape is the key

So how do you create these waves? Luckily it isn鈥檛 too hard. Random
electrical oscillations can begin spontaneously in almost any circuit: whether
they grow larger or fade away depends on variables such as the length of the
connections and the nature of its component parts. So a transistor that has been
carefully designed, with a specially tailored source and drain and a precisely
shaped gate contact, will begin to resonate with plasma waves at terahertz
frequencies. The precise frequency depends on the length of the transistor
channel and the bias voltage on the gate contact. This bias is crucial. It
controls the shape of the electron fluid inside the transistor鈥檚 channel, which
in turn changes the resonances that the device can support. Change the gate bias
and the resonance frequency will shift.

This is the key to using plasma-wave devices as detectors. Different
molecules absorb infrared light at different frequencies and to different
degrees. Shining infrared light with a broad spectrum of terahertz frequencies
through a sample of gas produces a 鈥渇ingerprint鈥� of missing bands鈥攁
spectrum鈥攚here the molecules have absorbed some of the frequencies. These
absorption spectra will indicate precisely what is lurking in the gas.

The device that Shur envisages will detect these molecular fingerprints by
the way they affect its plasma waves. Any light reaching the semiconductor
material is absorbed in the transistor channel and will superimpose its electric
field onto the bias voltage. This will excite plasma waves at specific
frequencies, which can be measured at the device鈥檚 output. 鈥淚t is just like a
radio receiver,鈥� says Shur. Changing the bias on the transistor鈥檚 gate can then
be used to 鈥渢une鈥� the detector to look at specific infrared bands. Better still,
scanning the bias voltage while measuring the transistor鈥檚 output turns the
sensor into a miniature terahertz spectrometer.

Shur鈥檚 first priority is to build one of these sensors. The plasma wave
detector should be 10 000 times as sensitive as existing high-frequency
detectors, he says. 鈥淭he devices would measure tiny, tiny concentrations of
different organic molecules. And it should be able to identify precisely what
these molecules are.鈥� The reason for this high sensitivity lies with the plasma
waves resonating inside the device. 鈥淚t鈥檚 like the system is in tune with an
incoming signal,鈥� he says. 鈥淛ust like sound reverberates in a well-designed
concert hall, these waves have a natural resonant cavity, which in this case is
the transistor itself.鈥� The slightest changes to these resonances will be easily
detected.

This should make these devices sensitive enough to detect just a few
molecules of explosives or poisonous substances hidden in a soup of other
chemicals. Such amazing sensitivity also makes the sensor ideal for studying our
environment, Shur suggests. It could pick out minute concentrations of
environmentally damaging chemicals, such as chlorine monoxide鈥攖he gas
mainly responsible for the catalytic destruction of ozone. Plasma-wave
transistors will also be extremely energy efficient, using only 1 per cent of
the energy required by existing high-frequency detectors. This would allow them
to be used in locations where power is at a premium.

Electronic flute

The detector will be followed by other plasma-wave devices, notably an
oscillator that would emit tunable infrared radiation. Changing the gate voltage
will change the frequency. Shur describes this as analogous to a flute鈥攊t
has only one input, but by pressing the valves shut, it will create an output at
many different frequencies. The transistors would radiate these signals via tiny
aerials.

Though none of these devices has yet been built, Shur and his team are
confident that they will work. They have already generated plasma waves in the
gigahertz range from nothing more sophisticated than a conventional
0.15-micrometre gallium nitride FET. The researchers simply cranked up the frequency
until it was way above its theoretical limit. 鈥淭he experimental results
confirmed that the detector operated at a frequency higher than the cut-off
frequency,鈥� says Lu. He believes that only plasma waves could have been
responsible for this signal. Better still, the researchers discovered that
changing the gate bias made the device respond just as their theories predicted
it would.

Shur, Lu and Dyakanov are working with Robert Weikle of the University of
Virginia to test the plasma-wave concept. A working device should be ready
within two years. They hope to use the nanofabrication facility at Cornell
University in Ithaca, New York, to make high electron mobility transistors
(HEMTs). These are high-speed transistors made from aluminium gallium arsenide
and gallium nitride, in which the conducting layer is only a few atoms thick.
Transistors with slimmed-down conducting channels can operate at much higher
frequencies than their fatter cousins. 鈥淭he gallium arsenide-based system
is the most promising at the moment,鈥� says Shur. Its electrons are almost ten
times as mobile as electrons in silicon, which means the device should work at
higher frequencies. Having tested the design with a single transistor, the next
step would be to build arrays of the devices on a chip to perform the binary
switching鈥攖he adding and multiplying work of a microprocessor.

In the longer term, Shur sees his technology achieving one of the key goals
of today鈥檚 electronics researchers: doing away with the wires that are now
needed to connect individual components on a chip. Instead they could be linked
through the ether by terahertz radiation, he says. These signals will behave
much like radio waves, but carry more information in a given time than a radio
signal ever could. 鈥淭he future is wireless,鈥� says Shur.

Radiation at terahertz frequencies has another useful property: it passes
straight through nonconducting material, including fabrics and many types of
building materials. Plastics, ceramics, explosives and drugs reflect or absorb
the radiation, which makes it a promising security and surveillance tool (see
鈥淣owhere to hide鈥�, New 杏吧原创 supplement, 4 November 1995, p 4).
Shur hopes that his terahertz wave detectors could be used to build imagers that
see through walls, or under rubble to locate disaster victims, or peer beneath
clothing to pick out hidden drugs or weapons.

Shur and his colleagues are certainly making waves in many of the places
where advanced electronics is discussed. The US Office of Naval Research is
considering whether to support RPI鈥檚 work. And Yue Kuo, a research scientist
specialising in thin film transistor fabrication at IBM鈥檚 Thomas J. Watson
Research Center in Yorktown Heights, New York, agrees that Shur鈥檚 work could
have great possibilities. It might one day offer terahertz chips operating at
speeds that are unimaginable with existing technology, he says.

Before that day dawns, the researchers will have to overcome some huge
challenges. At conventional frequencies, individual electronic components behave
as separate devices. But working at frequencies in the terahertz range,
plasma-wave devices become inseparable from the other components they are
connected to. One difficulty the team faces is how to feed terahertz radiation
into the transistor鈥檚 channel. Traditional designs may not be suitable.

Shur is undaunted by the formidable obstacles ahead. 鈥淣ew ideas are very
important,鈥� he says. Asked how he expects to analyse circuits at speeds where
there are鈥攁s yet鈥攏o analysis tools, he replies: 鈥淭hat鈥檚 a challenge.
But technology is always a challenge.鈥�