After a long wait, ferroelectric RAMs
head for volume production
MONTEREY, Calif. -- It's taken a decade, but ferroelectric
technology might finally be ready to make good on a promise to
rewrite the rules of the memory industry. Symetrix Corp., a
Colorado Springs, Colo., materials-research firm that for years
has championed the technology at the Integrated Ferroelectric
Symposium, reported at this year's sessions that volume
production of the parts is expected from at least three suppliers
before year's end.
First out of the gate will be Matsushita Electronics Corp., now
ramping one dedicated ferroelectric-RAM fab and building
another. Siemens AG, Hyundai, Motorola and others are
readying production of the memories. If all goes according to
plan, they would replace DRAMs and SRAMs in a host of
applications, effectively rewriting the rules of the memory
business and accounting for one-third of all semiconductor units
sold by the year 2010.
Separately,
in Japan,
Fujitsu
Ltd. and
NEC Corp.
have
announced
FeRAM
developments
based on
on a
zinc-zirconium-titanate
thin-film
compound
called
PZT,
which is more mature in its application base than Symetrix's
newer Y-1 thin film.
Non-volatile memories remain the centerpiece of interest for
ferroelectric thin films, which provide storage with virtually
unlimited read/writes. Papers at the symposium indicate the
broad areas in which the technology can be applied--as a
dielectric for very high-density DRAMs, as a thin film to help
reduce the size of RF components and as a piezoelectric sensor
film. But corporate activity last week centered very much on
memory applications.
For Symetrix chairman Carlos A. Paz de Araujo, it has been a
frustrating 15-year wait through conceiving the technology,
developing it at Symetrix and preaching its benefits to the world.
After more than a decade of being called "Dr. Cold Fusion" at
conferences, of having his work mistaken for ferromagnetics, of
having the company's Y-1 technology being misread as yttrium
iodine, he finally has his chance to change he world.
"Nobody believed, you know? Nobody believed. They
[Matsushita] had to double-prove, triple-prove it, within their
own company," Araujo said.
The promise of FeRAM is of a memory that can be written or
read extremely quickly, at a much lower voltage than E2PROMs
or SRAMs, without the eventual degradation of flash memory. In
other words, it outdistances the best possible features of each
memory type: high speed, low power, near-infinite durability.
Thanks to those kinds of sweeping promises, and the length of
time it's taken to bring FeRAM to market, more than a few
researchers roll their eyes at the mere mention of ferroelectrics.
But Araujo and believers like him seriously expect FeRAMs to
encroach on every aspect of electronics, steadily replacing
traditional memories as FeRAM densities climb. The parts are
being used only in embedded applications--particularly smart
cards--right now, but Araujo sees those parts making their way
into system memory as well, and Matsushita, through its
Panasonic brand name for consumer products, agrees.
"People say electronic money or [remote ID tags], but Panasonic
can find a great market in consumer electronics" for FeRAMs,
said Tatsuo Otsuki, general manager for Panasonic's IC-card
business office.
"There's a strong movement to use non-volatile memory as main
memory" in a PDA, Araujo said. His vision: an almost-infinitely
rewritable ferroelectric memory, which unlike SRAMs would
consume very little battery power.
At the heart of all these promises is the Y-1 compound.
Introduced in 1992, Y-1 is Symetrix's proprietary material, a
superlattice perovskite that Symetrix claims holds several
advantages over traditional ferroelectric materials such as PZT,
a compound of lead, zirconium and titanate. PZT is
more-susceptible to packaging problems, Araujo claimed, and
PZT only handles 107 rewrites, where Y-1 in the lab has
withstood 1014.
In addition to rewritability, Y-1 offers extremely fast read times,
rivaling those of fast SRAMs. But because the material can be
spread across extremely thin films without losing its electrical
properties, Y-1 parts operate at a fraction of SRAMs' voltage
needs, creating a serendipitous high-speed, low-power part.
Better still, Y-1 FeRAMs aren't expected to carry any price
premiums. At 50 cents apiece, the parts are expected to cost the
same as E2PROMs or even less.
Mass production
Matsushita is now in volume production with FeRAM-based
smart cards, churning them out of a 6-inch fab in Japan that will
reach its 20,000-wafer-per-month maximum output some time
around October. That fab uses 0.6-micron process rules; an
8-inch, 0.35-micron fab also being built in Japan will be wholly
dedicated to FeRAMs, reaching full production and a migration
to 0.25-micron rules around 2000.
In Matsushita's case, FeRAM parts have been four years
coming, as the company struggled with such technical hurdles as
the difficulty of packaging the parts in plastic, which has been a
barrier to ferroelectric production in general, Otsuki said.
Matsushita happens to be the first, but it's not the only FeRAM
believer. Siemens is constructing a fab in Germany dedicated to
ferroelectric parts, and Hyundai is preparing an enormous facility
to build the parts. Of the 10 chip makers licensed by Symetrix,
two besides Panasonic plan to announce parts this year, Araujo
said.
Matsushita's first production of FeRAMs is targeted at
embedded applications, particularly smart cards, where
ferroelectric technology has "found its real calling," Araujo said.
Possibilities here include contactless ID tags, which can be read
and overwritten as a person walks past a reader, or the holy grail
of embedded technology: bar- code replacement.
Matsushita's later plans include a standalone 64-kbit FeRAM to
roll out in the fall.
Hurdles remain before FeRAMs can broadly replace DRAMs.
Memory experts in Japan have discovered that, to guarantee
stable operation, FeRAMs usually must be based on a 2T/2C
structure-two transistors and two ferroelectric capacitors. That
cell size is theoretically twice as large as DRAM. Some
researchers have proposed one-transistor and
one-ferroelectric-capacitor (1T/1C) as the structure with the
greatest potential for high density, but bottlenecks in
implementation can hinder stable operation.
To increase FeRAM density, a small cell area and a high stable
operation are both essential.
Fujitsu pursued the 1T/1C structure, which it prefers to call a 1T
structure. As the cell occupies the space of a single transistor, it
can be made as small as that of DRAM or much smaller. Based
on the theoretically smallest form, the Fujitsu research team
tried to improve stability in operation.
NEC, on the other hand, took the golden mean, a one-transistor
and two-ferroelectric-capacitor (1T/2C) structure implemented
in PZT. It was said to have a stable operation and its cell size
can be made as small as that of DRAM.
In 1T/2C structures, of course, the two capacitors would need
more space than one capacitor of the same size in the 1T/1C
structure. But since the reference voltage is fixed, the capacitors
are required to generate only a small voltage, which can be
differentiated by the sense amplifier-around 70 mV to 100 mV
depending on the sensitivity of the amplifier. "This makes it
possible to make the capacitor size smaller than the capacitor of
an unstable 1T/1C cell," said Hiromitsu Hada, manager of NEC's
ultralarge-scale-integration (ULSI) laboratory.
NEC fabricated the cell structure with currently available
materials and processes, which resulted in a larger cell than the
theoretical size cell. But the researchers verified the cell's
working stability. They also simulated that the cell operation, in
the equivalent size to that of 256-Mbit DRAM, is stable enough,
Hada said. NEC concluded that the 1T/2C structure is promising
for high-density FeRAMs.
To follow up, NEC researchers are now improving materials and
processes and studying the cell layout to make it smaller. "By
improving material to reduce 'fatigue,' we want to increase the
number of rewritable times from around 106 to 1,014," Hada said.
"For applications which do not need large density such as IC
cards, NEC has already established the FeRAM technology,"
said Hada. "FeRAM has quite a potential to substitute for
DRAMs, but there is still a lot to do and it needs breakthroughs
for such high-density applications."
Since FeRAM employs metal-oxide materials, it can be difficult
to merge the thin-film processes with CMOS processes, based
on reduction gases. As feature sizes become finer, the impact of
undesirable deoxidization becomes greater. Fujitsu addressed
that problem by separating CMOS and ferroelectric processes,
which carries a possible fringe benefit of reducing size and
increasing stability in operation.
Explaining why Fujitsu focused on a small, simple structure,
Yoshihiro Arimoto, general manager at Fujitsu's fundamental
laboratory, said that "to establish stability with a small cell
structure is a challenge. But how many chips can be cut out from
one wafer is very important." He said that FeRAM will begin to
replace DRAMs and flash memories "when the cost of FeRAM
starts to drop. But at present, we are at the stage that its
operation has just been simulated."
Fujitsu's 1T cell was fabricated by adding to the conventional
CMOS process a planar ferroelectric film and a single electrode
layer. The Pt electrode layer was turned into pairs of electrodes
to form coplanar electrodes.
"We developed the structure so that it goes well with a CMOS
process. After a CMOS transistor is fabricated, the ferroelectric
process follows," said Masaki Aoki, the researcher in charge of
FeRAM development in Fujitsu laboratories' ULSI process
department.
Since the structure differs from a conventional 1T/1C, Fujitsu's
cell no longer suffers from the instability peculiar to the 1T/1C,
researchers said.
Fujitsu improved the structure of 1T, which had been proposed
years ago. By contrast with a 2T/2C structure, which needs large
polarization, the 1T structure needs only a small charge, which is
suitable for a coplanar ferroelectric capacitor that has electrodes
on the same plane.
This structure makes the capacitor compact compared with the
structure in which electrodes sandwich the capacitor, according
to the researchers.
Furthermore, read-out in the 1T structure can be done with about
one-fifth of a writing voltage, which does not change the status
of the cell. "The data is retained after read-out operations, which
contributes high reliability and low power consumption," said
Takashi Eshita, chief researcher in the ULSI process
department. |
