快猫短视频

TAKE some Cold War research behind the Iron Curtain, add a love affair
between an American physicist and a Russian and mix in a hi-tech start-up
company with global ambitions. This is not, however, a recently discovered James
Bond novel. This is the story of how ferroelectric memory arrived, this week, on
the world scene.

Over the past few days, motorists calling at Shell petrol stations across
Japan have been given a smart card that they can charge up with money. For them,
there鈥檚 no more queueing to pay, no more swiping, no more waiting for
authorisation. They just wave their cards at the pump, key in a PIN number, and
off they go. The cards communicate by radio and have ferroelectric memories that
operate up to a million times faster than their rivals. With these cards, a
revolution in chip making has finally arrived.

Transparent and yellowish, like dirty glass, ferroelectric thin films are
made from an obscure family of crystals called perovskites. They may not look
much, but the films possess extraordinary properties when it comes to storing
electric charge. They can pack away far more charge than the materials used
today in silicon chips鈥攚ithout ferroelectric capacitors, mobile phones
would never have shrunk as small as they have. And the charge within the films
can be switched quickly between two states, both of which are stable even when
the power is turned off. This is ideal behaviour for making nonvolatile computer
memory; one day ferroelectric chips could replace both the hard disc and the RAM
in your computer.

For ferroelectrics, success has been a long time coming. Researchers first
proposed them as high-speed memories back in the early 1950s. Many American
electronics companies, including IBM and Bell Laboratories, took part in this
research. But eventually they gave up because the technology of the time was
simply too primitive to create defect-free thin films.

During the 1970s, Russians made much of the running in ferroelectric
research. Preeminent among them was G. A. Smolensky, a physicist at the Ioffe
Institute in Leningrad. Today鈥檚 trend towards ferroelectric films can be traced
back to his work and the chance arrival in Russia of an American physicist.

In 1981, Jim Scott, formerly of Bell Labs, went to work with Smolensky on
perovskites, and during his year there married a Russian. Three years later,
back at the University of Colorado in Boulder, Scott got a call out of the blue
from Carlos Araujo, director of the microelectronics centre at the university鈥檚
campus in Colorado Springs.

Araujo wanted to take advantage of new thin-film fabrication techniques to
make ferroelectric memories and needed a translation of a book by another
Russian, Yuri Tomashpolski. By incredible good luck, the ideal translation team,
ferroelectric specialist Scott and his wife Galina, lived only two hours away.
Scott recalls asking: 鈥淗ow long is the book, and how soon do you have to have
it?鈥� Araujo replied: 鈥淲ell, it鈥檚 471 pages, and we鈥檇 like it by Tuesday.鈥�

From this initial contact a fruitful collaboration developed between
physicist and engineer. In 1986, together with Larry McMillan, a former
engineering director at Motorola, they formed a company called Symetrix to
research ferroelectric materials. It was not long before the company began to
attract attention from Japan.

In March 1991, the giant Matsushita Electronics paid Symetrix $1
million to solve a miniaturisation problem. The size of its mobile phones was
limited by a microwave-frequency amplifier chip, 90 per cent of which was taken
up by a single capacitor. The Symetrix researchers decided to replace the
outsize capacitor with one made with a ferroelectric film of barium strontium
titanate (BST).

Scary stuff

The rationale was simple. The perovskite has a dielectric constant鈥攁
measure of how much charge it can store鈥攁bout a hundred times that of more
usual dielectric materials, such as silicon dioxide and silicon nitride. If
Matsushita could increase the dielectric constant of its capacitor a
hundredfold, it could decrease the capacitor鈥檚 area by the same factor. As Scott
points out, 鈥渇actors of a hundred are pretty serious, even among friends鈥�.

The big problem facing Symetrix was that to conservative semiconductor
process engineers, the idea of introducing an exotic material such as BST in
their precious wafer production lines was scary stuff. The Symetrix researchers
had a solution. They found a way to add the thin film as a 鈥渂ack-end鈥� process in
a separate production line after the silicon or gallium arsenide structures had
been etched.

The ferroelectric chip that Symetrix designed is 50 times smaller than the
one it replaced. Matsushita put the device into high volume production. Last
year, the Japanese firm cranked out 60 million ferroelectric chips, which are
now used in 75 per cent of digital mobile phones worldwide.

The ability to mass-produce ferroelectric capacitors also promises good news
for the makers of memory chips. The dynamic random access memory (DRAM) that
forms the main memory in PCs, remembers data by storing charge in capacitors. As
demand for memory grows so does the density of capacitors on the chips. For
several generations now, manufacturers have tried to increase the density by,
for example, stacking capacitors on top of one another. But even such elaborate
methods are nearing their limits.

At the present rate of advance, 4 gigabit chips should arrive around 2004.
But this won鈥檛 happen, says Scott, if chipmakers stick to capacitors with low
dielectric constants. 鈥淲hatever objections you have to using exotic materials
like ferroelectrics, you have no choice,鈥� he argues. 鈥淭he consensus for DRAMs is
coming down to barium strontium titanate.鈥�

Loosely bound

Like other ferroelectrics, BST can store large amounts of charge because of
the unusual nature of its crystals. These are made up of simple box-like
structures with barium atoms at the corners and oxygens in the centre of each
face. In the centre of the box sits a single atom of strontium or titanium. With
a normal dielectric, an applied voltage simply polarises charge within the
material鈥檚 atoms. But in BST the central atom is only loosely bound and can
move, giving added scope for polarisation.

But the strange properties of ferroelectrics do not end there. In some
crystals, notably lead zirconium titanate (PZT), which has been the focus of
much research, the central atom has two stable positions鈥攐ne above centre
and one below. A voltage applied across the crystal, first one way then the
other, will flip the atom from one state to the other. The crystal鈥檚 polarity
follows suit, and if the voltage is switched off the crystal stays in the last
state it was in
(see Diagram). These abilities promised to make
ferroelectric films valuable memories in their own right鈥攂ut for one
snag.

To read which state a crystal鈥檚 central atom is in (whether it holds a binary
1 or 0) you would apply, say, a positive voltage across the crystal. If the
crystal is already positively polarised, nothing much happens鈥攑erhaps a
tiny transient current flows. But if the crystal is negatively polarised then
its central atom and the crystals charge distribution flip positions and there
is a large transient current. This tells you which 鈥渂it鈥� the crystal stored, but
it also destroys that information. So every time information is read, it must be
written back into memory again.

The snag is that materials such as PZT get fatigued. For each rewrite, the
amount of stored charge decreases until ultimately it becomes unreadable. PZT
reaches this point at around a billion cycles, well below the trillion level
(10¹²) needed for commercial devices.

Fatigue was the next problem Symetrix tackled. Oxygen exists in perovskites
as ions with two extra electrons, and over time these are attracted to the
positive anode. 鈥淲hen oxygens migrate through the crystal, they leave behind
vacancies which are filled by electrons,鈥� says Scott. And electrons are in
plentiful supply: they tend to form a 鈥渃loud鈥� near where they enter the crystal
at the cathode. But when electrons replace oxygen ions, they leave charge
defects in the crystal lattice that set up their own electric fields.
Eventually, these local fields can become so large that they cancel out the
field produced by an applied voltage.

Researchers studying PZT tried to compensate for this oxygen depletion by
using electrodes rich in oxygen. But the electrodes leaked electric current.
Symetrix researchers decided to attack the problem from the other direction, by
putting movable oxygen atoms into the ferroelectric itself. They abandoned
conventional perovskites for a more complex family of crystals. These, too, had
a box-like crystalline structure, but came with additional bismuth oxide planes
slotted in between every two or three or four boxes
(see Diagram).

Their intuition proved to be correct. In their chosen material, strontium
bismuth tantalate, which they labelled Y-1, the extra electrons are mopped up.
In tests they found that even after a trillion cycles, there was no degradation
in the charge stored by the film. Araujo and his colleagues had shown for the
first time that it was possible to totally eliminate fatigue.

The initial reaction to Y-1 was disbelief. 鈥淲e had to fight the other
interests in ferroelectrics based on PZT,鈥� Araujo recalls. 鈥淭hey were fifteen
times better funded and had twenty times more people.鈥�

Today, things are different. Symetrix has issued licenses for Y-1 to some 14
companies, including giants such as NEC, Mitsubishi and Toshiba in Japan,
Motorola in the US, Siemens in Germany and Hyundai in Korea. The rush to sign up
is driven by the advantages that ferroelectrics offer over other nonvolatile
memories鈥攅lectrically erasable programmable read-only memory (EEPROM) and
flash memory. Ferroelectric memories are more durable and operate at lower
power. The fact that they are easy to manufacture鈥攋ust three steps on top
of the standard process for making silicon devices鈥攕hould also make them
cheaper.

But by far the biggest plus is the speed at which information can be written
to them. A ferroelectric memory can be erased and rewritten at least a thousand,
and possibly as much as a million times faster than an EEPROM.

But although Y-1 promised much and was in great demand, Symetrix was not yet
out of the woods. It took another three years to include prototype ferroelectric
films into robust commercial devices. The chief problem was 鈥渋mprint鈥�, a
consequence of the high-temperature tests that manufacturers put their chips
through. First, they heat the chips up to about 150 掳C, then they zap them
with an electric pulse several billion times. They switch the state of every
memory cell in turn鈥攊f it was a 1, they write a 0, and vice versa. Testing
shows whether the correct state is registered. With their Y-1 devices, the
Symetrix researchers found all too often that the memory was imprinted with all
0s or all 1s: in other words, completely useless.

The thinner the ferroelectric film鈥攖ypically about 200
nanometres鈥攖he more susceptible it is to imprint. In order to overcome the
problem, the researchers had to change the ratios of the different elements in
the crystals until they found an optimum. 鈥淲hen we solved the fatigue problem,
we were all happy because we thought we were done,鈥� recalls Araujo, 鈥渂ut then we
had to solve imprint, and that was much tougher than fatigue. Until we figured
out how to fix it there was really no technology, all we had was poetry.鈥�

With Symetrix鈥檚 help, Matsushita is now ahead of its competitors and is
launching ferroelectric memories in three markets. The first is the contactless
smart cards launched this week. In addition to a tiny chip, 4 millimetres
square, each card has a radio antenna built into the edge of the plastic. Radio
signals can read the card鈥檚 memory and write to it from about 2 metres away. The
card gets enough power to reply from the reader鈥檚 signal, says McMillan.

In addition to paying for petrol or road tolls, this has many other potential
applications. An obvious one is as season tickets for commuters. When Japanese
train stations switched to automated ticket gates a few years ago, commuters
were annoyed because instead of flashing their passes at ticket collectors, they
had to remove the tickets from their wallets in order to put them in the gates.
The new system allows commuters to wave their passes over a reader.

An offshoot of smart cards will be radio-frequency ID tags. These will
substitute for bar codes in many applications, most notably as baggage tags used
at airport check-ins. But the tags will have to be cheap, 50 cents or less. In
the short term, there is not enough demand for tags to feed the ravening maw of
a semiconductor plant. So tags will probably have wait until they can piggyback
on mainstream ferroelectric products such as smart cards.

Matsushita鈥檚 second product is a 64-kilobit memory embedded in an 8-bit
silicon microprocessor. Chip makers ship about 3.5 billion such devices a year
for use as microcontrollers. There are about 100 microcontrollers in every home
in the washing machine, microwave oven, telephone and so on, and about 20 more
in an average car.

Finally, there is the stand-alone memory chip. Here, Matsushita has developed
a 256-kilobit chip as a much cheaper substitute for static random access memory
(SRAM), which needs a battery to prevent stored data from disappearing. Such
memories are used, for example, in videotape recorders to keep track of the time
when the power is switched off.

Companies such as Hyundai and Siemens are also racing to develop 4-megabit
ferroelectric memories. The attraction here is handheld computers that connect
to a network. Since most of their data is stored remotely, such machines need
only about 8 megabytes of memory, and no hard disc drive.

If they came with DRAM and nonvolatile memory, these devices would eat
batteries and cost too much. So the obvious answer is to use just nonvolatile
memory, says Araujo. But all the candidates are too slow or suffer fatigue and,
up to now, large ferroelectric memory chips have been confined to labs. 鈥淏ut the
minute one comes out,鈥� and Araujo predicts that Siemens will launch such a chip
early next year, 鈥渢hen you鈥檒l see a stampede.鈥�