Milwaukee, Wisconsin
HERE鈥橲 a dilemma: you鈥檙e at a party and are handed a glass of wine of
uncertain origin. You鈥檝e a delicate palette and don鈥檛 want to risk scouring your
tongue with a swig of something the French wouldn鈥檛 wash their cars with. The
solution? Discretely whip out your hand-held electronic nose and give it a heady
sniff or two of your glass. Seconds later, if the green light flicks on, it鈥檚 OK
to quaff. But if the red light flashes, find a convenient plant pot.
Portable pre-programmed connoisseurs may be a little way off yet, but
scientists are nevertheless on the scent of something big鈥攖hey reckon they
can match the capabilities of a human or animal nose using a bunch of optical
fibres or a microchip. Artificial olfaction may seem a tall order, but simple
sniffers have been around for a while. Electronic noses first hit the market
about five years ago. Based on arrays of electrochemical sensors connected to a
PC, they can pass or fail batches of beer, aeroplane de-icer, and other smelly
products, having been trained to recognise the chemicals present in batches
rejected by human noses. Now scientists think they can take electronic noses
into the realms of wine buffs and even sniffer dogs.
Advertisement
Some researchers believe that the arrival of electronic noses, shrunk from
today鈥檚 benchtop devices to the size of a single silicon chip, will herald a
revolution in artificial olfaction. These integrated 鈥渘ose brains鈥濃�攕mall,
powerful, and consuming only a whiff of battery power鈥攃ould become as
ubiquitous as the microprocessor, finding their way into telephones, ovens, cars
and hospital wards. 鈥淧utting it on a chip allows you to apply it in many, many
different ways that you could never do with a benchtop instrument,鈥� says
materials chemist Nathan Lewis, who is developing a nose chip at the California
Institute of Technology (Caltech) in Pasadena. Along the way, he and other
scientists also hope to resolve some of the mysteries of natural olfaction.
For all these lofty aspirations, after spending decades developing electronic
noses, George Dodd, who could be described as the father of the technology,
still looks upon sniffer dogs with envy. These veritable 鈥渘oses with legs鈥� can
sniff out the exceedingly faint odour of buried land mines, find drugs secreted
in airline luggage, or track the body odour of a fugitive across miles of smelly
countryside. But the olfactory apparatus of the sniffer dog is as paradoxical as
it is amazing, says Dodd, a senior research fellow at the Highlands Scientific
Research Group in Inverness, Scotland.
Think of a sniffer dog as a radio receiver that, rather than tuning into
electromagnetic waves, picks up the odour molecules that leak into the air from
petrol, perfume, garlic oil and other smelly substances. It can detect a vast
spectrum of odours with crystal clarity, even when they consist of only a
handful of molecules diluted in billions of molecules of air. Radio engineers
would be hard pressed to match such a feat. 鈥淎 dog鈥檚 sense of smell has ultimate
bandwidth,鈥� Dodd says. 鈥淚t can smell everything, but it can also do it with
fantastic sensitivity, and that鈥檚 paradoxical.鈥�
Swelled with success
In the early 1980s, Dodd and colleagues at the University of Warwick made the
first breakthrough in artificial smelling when they found a way to mimic one
characteristic of the dog鈥檚 nose鈥攊ts broad sensitivity to odours. They
constructed an array of three chemical sensors made from different conducting
polymers鈥攑lastics that conduct electricity. Internally, on a microscopic
scale, the materials look like a tangled ball of spaghetti. When they absorb
odour molecules, the polymers swell. This changes their conductivity and causes
a measurable change in the size of an electrical current being passed through
them.
Each of Dodd鈥檚 three plastic sensors responded differently to odour
molecules. When presented with, say, the scent of garlic, sensors one and two
might strongly absorb the odour and give a bigger drop in conductivity than
sensor three. But when exposed to rose oil, the sensors鈥� pattern of response
would be entirely different, perhaps with only sensor three strongly absorbing
the scent. This gave a unique electrical fingerprint for each odour sniffed.
Dodd鈥檚 device was the first step towards the electronic nose.
Today鈥檚 commercial noses work on the same principle, but contain more sensors
to give a wider variety of responses. The noses use neural networks to learn the
patterns of responses associated with different smells. Before they are put to
work they must be trained to recognise certain smells, such as the chemicals
which are characteristic of a bad batch of beer. At one stage of the brewing
process, for example, a buttery 鈥渙ff鈥� flavour can develop. This is due to the
presence of diacetyl, which is converted to tasteless butane-diol by resting the
batch for a while. The nose is exposed to batches of beer both with and without
diacetyl, and told each time whether the batches are off, or OK. After repeated
exposure to these smells, the nose can tell them apart very accurately, and can
then be put to work sniffing out diacetyl (Technology, 20 February 1993, p
18).
Neotronics, a British nose manufacturer based at Bishop鈥檚 Stortford in
Hertfordshire, has put more than 100 units to work in North America and Europe.
Each contains 12 sensors which can be trained to perform a multitude of tasks.
They are mostly used for routine quality control in the food and beverage
industry, where they replace the subjective and sometimes unreliable sniffing
and sipping of human 鈥渙dour panels鈥�.
But researchers are already working on the next generation of noses, taking
the technology much closer to the real thing. Neuroscientist John Kauer at Tufts
University Medical School in Boston, and chemist David Walt at Tufts University
in Medford, Massachusetts, think the best way to improve the technology is to
copy what nature has to offer. 鈥淲hat we鈥檙e trying to do is learn from the
biological system and apply those principles to design an artificial system that
takes advantage of all those features,鈥� Walt says.
This explains why the disembodied heads of a dog, a goat and several rats
stare mutely from benches in Kauer鈥檚 laboratory. He is using them as templates
for constructing artificial nasal cavities to see if airflow patterns over
sensors have anything to do with sensitivity. Dogs have more intricately folded
cavities than humans, and these may set up airflow patterns that contribute to
their remarkable olfactory powers. The team are only at the modelling stage at
present, but future noses, Kauer speculates, might use a specially moulded
鈥渘asal cavity鈥� studded with chemical sensors. 鈥淲ho knows,鈥� he says. 鈥淸Cavity
shape] may enhance the sensitivity in ways we don鈥檛 yet understand.鈥�
Pea-sized organ
In humans, the hardware of smell resides high in the nasal passages in a
patch of tissue called the nasal epithelium. As smells waft over it, they
encounter an estimated 10 million sensory cells. Each is equipped with perhaps
as many as a thousand different types of chemical receptors. With each sniff,
odour molecules dock with a particular pattern of receptors and trigger a
variety of electrical signals that are sent to a pair of pea-sized organs at the
base of the brain called the olfactory bulbs. Here they are sorted and sent for
processing.
Humans can detect a wide spectrum of odours as well as the subtle shades
between them, albeit not as well as dogs. Babies only a week old can identify
their mother by her smell, just as a dog can recognise the genetically
determined body odour of its master. How they do it, Kauer says, comes down to
the complexity of the information that the nose sends to the brain with each
sniff.
Dyed in the nose
To mimic the nasal epithelium and its high sensitivity, the Tufts nose uses a
sheath of optical fibres. Walt coats the tips of the fibres with various
plastics mixed with a dye called Nile Red, and these act as sensors by capturing
the odour molecules. Light is shone down the fibres to the tips, where it makes
the dye fluoresce. Part of that fluorescent light travels back up the fibre,
where its intensity is measured by a charge-coupled device鈥攁n 鈥渆lectronic
eye鈥濃�攁nd turned into digital signals for analysis by a neural network. As
in Neotronics鈥� commercial electronic noses, the plastics absorb odour molecules
and swell to different degrees. This changes the intensity of light going back
up the fibre.
Walt and Kauer believe their fibre-optic sniffer has major advantages over
existing commercial noses. Most important is the richness of information that
fibre-optic sensors provide. Conducting polymer sensors, Walt says, measure a
single variable, namely the change in conductivity when they soak up smells. 鈥淎t
some point you are going to exhaust the diversity of these systems simply
because all the sensors are intrinsically alike,鈥� he thinks. In the Tufts
optical fibre system, on the other hand, when a plastic captures odour molecules
both the intensity and wavelength of light it sends back up the fibre change, as
does the fluorescence lifetime of the dye鈥攖he time it takes for the dye
molecules to emit light depends on the type and number of odour molecules
surrounding them.
鈥淪o in the very simplest case you have at least three different parameters
that you can measure with an optical device,鈥� Kauer says. He believes that this
greater complexity enhances the power of the device to discriminate between
smells.
In recent tests, the Tufts nose distinguished between three types of alcohol
in which the structures differed by only a single carbon atom. More
impressively, the nose identified the molecules in a mixture of two chemicals,
and the team thinks it will also be able to tell how much of each is in the
mixture. This is a task that today鈥檚 nose technology falls short on, says Walt.
鈥淲hile the existing systems may be able to tell you that a wine is the same or
different from what it was trained on, our system should be able to tell you how
it differs, what the components are in each of those different wines, and how
they differ in a quantitative fashion.鈥�
However, judging the researchers latest gadgets against commercial sniffers
is a 鈥渂ogus comparison鈥�, says Ley Hathcock, director of research for Neotronics
Scientific, the company鈥檚 electronic nose division. Commercial noses are
designed to excel at quality control, determining whether today鈥檚 batch of beer
matches yesterday鈥檚. 鈥淚t鈥檚 an instrument that sits in a laboratory or factory
floor that does a damn good job at quality control,鈥� Hathcock says. 鈥淲hat it鈥檚
not is an artificial dog鈥檚 nose.鈥�
But suppose you want an artificial dog鈥檚 nose? The Tufts optical system could
use a neural network to process the signals. But at Caltech, Lewis has come up
with a different way to do it: build an entire olfactory system on a single
silicon chip. To create this ultimate 鈥渘ose brain鈥�, he is relying on a similar
concept to the one that has made the Neotronics noses such a success鈥攂ut
on a much grander scale. Lewis intends to install 10 000 or more plastic sensors
on a chip along with all the neural network circuitry to run them.
鈥淭he first thing you want to do, whether the nose is on a chip or not, is to
create a diverse library of sensors,鈥� Lewis says. 鈥淵ou want the sensors to be as
broadly probing of what is out there as possible with as many sensors that are
incrementally different as you can make.鈥� Such an array would allow a nose chip
to gather a massive number of responses to odour molecules, increasing its
ability to distinguish between smells.
But this approach is controversial. Other researchers believe that it is the
degree of difference between the sensors, not the overall number, that is most
important. They suggest that the same level of information can be gleaned using
10 sensors as with a million. 鈥淲e disagree with that,鈥� says Lewis.
He admits that using a minimum of sensors is appropriate for tailoring
electronic noses to specific applications. 鈥淚f you want to use this to smell
smoke in your house and you can do that with 10 sensors, then you should do that
with 10 sensors,鈥� Lewis says. But when it comes to mimicking natural olfaction,
he believes, the more information you can gather from 鈥渙dour space鈥� the
better.
Like the Neotronics noses, Lewis鈥檚 device relies on sensors made from
plastics. But the key difference, which gives Lewis a much wider range of
materials to play with, is that he can use any plastic he can get his hands on,
rather than being limited to conductive plastics (see 鈥淩ecipes for
蝉别苍蝉辞谤蝉鈥�).
Last year, the Caltech team tested a benchtop nose with an array of 17
sensors, and it could easily tell the difference between wine and spirits or a
fresh fish from a rotting one鈥攁lbeit rather slowly for the olfactory
comfort of the researchers. Since then, Lewis鈥檚 nose has started shrinking. He
and electrical engineer Rod Goodman have fashioned a demonstration device
comprising a silicon chip with five sensors. They etched tiny wells, about a
millimetre wide, on a 1-centimetre wide slab of silicon and ran two printed
circuit wires to each well. Then they deposited a different type of sensor
material in each.
No brain
The nose worked. 鈥淵ou can breathe on it and it shows a pattern,鈥� says Lewis
proudly. But lacking the circuitry to make sense of the molecules, it is a nose
without a brain. He and Goodman, whose speciality is designing chips that use
neural networks, are now working to get the brains onboard. Within three years,
they hope to have 10 000 sensors and the necessary neural network hardware on a
single chip. This is not the stuff of science fiction, Lewis emphasises. The
basic technology to put a nose on a chip is already available. He believes the
neural network needed to run the nose should be no more advanced than those in
portable computers that learn to recognise handwriting. 鈥淚t鈥檚 not hard,鈥� he
says. 鈥淚t鈥檚 just not been done.鈥�
So what could you do with a nose on a chip? The simple answer is, anything
you want. Small mass-produced noses could crop up anywhere, provided they are
cheap, simple and battery-powered. 鈥淸You could] put one in your toaster, in your
microwave oven, on your wall for a smoke detector, or in your car to sense 15
different things,鈥� says Lewis.
Today鈥檚 commercial noses take one to two minutes to fingerprint a smell
because they need to get a steady reading from the sensors at a controlled
temperature, but a nose chip would be faster. For one thing, the sensors and
neural circuitry would all be so close together that delays would be minimal.
Also, smell molecules would take less time to swell the plastic in smaller
sensors. Without a fast nose, Lewis says, 鈥測ou can鈥檛 think about tracking an
odour to its source, allowing for changing wind direction. You can鈥檛 think about
putting it on the nose of a robot and have it go find the bomb in the airport.
All those things need rapid response, in real time, in a challenging
environment. That鈥檚 what we want to open up.鈥�
Dodd also sees the nose on a chip as a significant advance. In fact, he and
John Barker, a professor of electronics at Glasgow University, are planning to
build one. They want to use the new noses to diagnose diseases from odour
molecules in patients鈥� breath. Candidate diseases Dodd and other electronic nose
researchers are interested in include cirrhosis of the liver, diabetes, lung
cancer, and ulcers.
Dodd thinks liver disease is a particularly promising target. 鈥淲hen people
have liver disease they have a peculiar smell about them,鈥� he says. 鈥淭hat鈥檚 very
well known in medicine.鈥� The first step is to identify the diagnostic odour
molecules. Then the sensor array on the nose chip could be engineered to be
highly sensitive to these key smells.
Bad breath
Electronic noses could bring about a revolution in telemedicine. A patient
could breathe into a phone equipped with a nose chip, Dodd says, which would
send the odour information over the phone lines to doctors. 鈥淚f we can identify
a small number of molecules in liver cirrhosis we can make the chip so sensitive
I reckon we may be able to diagnose liver cirrhosis using telemedicine before
anybody would ever dream of going to a doctor,鈥� he says.
As Dodd and other scientists of smell dream of what they would do with better
electronic noses, they do so with an eye on the mysteries of olfaction itself.
Kauer believes the process of inventing the next generation of electronic noses
may spawn new insights into biology. 鈥淚t forces you to think really carefully
and critically about what you think you understand and do not understand about a
system,鈥� he says. 鈥淎nd if you can understand it well enough to build it, you鈥檝e
come a long way.鈥�
Lewis, like Kauer, is in the electronic nose business mainly for intellectual
rather than financial reward. Telemedicine and other futuristic applications
that his nose on a chip may bring are the gravy, not the meat, of the
enterprise. 鈥淲e want to try to implement the biological system so that we can
learn,鈥� he says. 鈥淚f along the way we learn how to build a better nose gizmo
that goes in your toaster, then that鈥檚 fine.鈥�
* * *
Recipes for sensors
TO MAKE a nose chip with 10 000 sensors, you need to find a way to cook up a
large number of sensors that each respond slightly differently to odours. Nathan
Lewis and his colleagues at Caltech have developed a surprisingly simple way to
do it鈥攕o simple, in fact, that Lewis has offered it as a science fair
project for teenagers. The researchers use commercially available plastics to
make their sensors. Viewed under a microscope, these plastics are a tangled mass
of fibres, which swell when odour molecules are absorbed in-between the strands.
The Caltech team has experimented with polystyrene, which is used to make
disposable coffee cups, and polyvinyl alcohol, which is used in plastic food
wrapping.
Particles of carbon black, an electrical conductor found in pencil lead and
car tyres, are then scattered among the strands of the polymers. Because the
polymers are non-conducting, electric current can flow through them only where
there is a pathway formed by carbon black particles. When odour molecules make
the polymers swell, the particles are spread gradually farther apart, breaking
up the pathways and so reducing the conductivity of the material. Virtually any
plastic can be used. 鈥淭hat鈥檚 the beauty of it,鈥� Lewis says gleefully. 鈥淲e take
anything we can buy, sprinkle in a little bit of conductor, and we鈥檙e off to the
谤补肠别蝉.鈥�
Different plastic sensors swell to different degrees depending on their
chemical affinities. Polyvinyl alcohol, for example, is hydrophilic. A sensor
which is made from this material will readily soak up the watery odour of wine,
say, causing conductivity to fall markedly. In contrast, polystyrene is
hydrophobic. Exposed to the same wine, it would not swell as much as the
polyvinyl alcohol, and its conductivity would drop less.
In addition to using different plastics, the cooks at Caltech can create an
even greater variety of sensors by varying the amount of carbon black they fold
in. Two sensors of the same plastic, one with 10 per cent by weight of carbon
black, the other with 20 per cent, would swell by the same amount when exposed
to a particular odour, but the conductivity of the sensor with less carbon black
would decrease by more, because more of the conducting pathways would be broken.
By mixing polymers with varying amounts of carbon black, Lewis thinks he can
make the large number of sensors he would need for his nose chip.
Is there a limit to the number of different sensors they can make this way?
鈥淗ow many different types of plastics can you buy?鈥� says Lewis.