[AI] Five ways to revolutionise computer memory

Sanjay ilovecold at gmail.com
Tue Apr 20 02:39:23 EDT 2010

          You can store all your music on a personal MP3 player - which
          technology will do the same for your high-definition movie

by Joerg Heber

Once upon a time, not so long ago, the idea that you might store your
entire music collection on a single hand-held device would have been
greeted with disbelief. Ditto backing up all your essential computer
files using a memory stick key ring, or storing thousands of
high-resolution holiday snaps in one pocket-sized camera.

What a difference a decade makes. The impossible has become possible
thanks to the lightning rise of a memory technology with the
snazzy name of "flash".

So where is the technology that can store our high-definition home
cinema collection on a single chip? Or every book we would ever want
to read or refer to? Flash can't do that. In labs across the world,
though, an impressive array of technologies is lining up that could
make such dreams achievable.

These "supermemories" are close to realising a vision set out by
revered physicist Richard Feynman 50 years ago this month. In a
lecture to the American Physical Society entitled "There's plenty
of room at the bottom", he asked whether it might ever be possible to
write all 24 volumes of the Encyclopaedia Britannica on a pin head.
Each tiny ink dot used to print each letter would have to be reduced
to the size of just 1000 atoms, he calculated - a square with sides of
just 9 nanometres.

Feynman speculated that people looking back from the year 2000 would
wonder why it took till 1960 before we began to explore this "room at
the bottom" - what we now know as the nanoscale. Late start or not,
the progress in miniaturising information storage in the decades since
has been stunning. Today, the smallest feature that can store a bit of
information is some 40 nanometres across in commercial flash devices.
The first flash chips capable of storing 64 gigabits of
information were shipped just a couple of months ago.

The kinds of technologies Feynman was talking about, though, would fit
terabytes of data on a single chip. That requires a design simpler
even than the already admirably straightforward flash architecture
(see "Flash: memory hero"). The mechanism for reading and writing
the memory would also have to be reliable and, above all, fast, taking
just nanoseconds. And the memory should be stable: once written, it
should not degrade for at least a decade.

That is quite a shopping list. Whatever technology fits the bill will
not be flash, but it will be mightily impressive. It won't be easy
establishing it, with flash already so well entrenched, but with the
market for memory chips worth something between $20 billion and $30
billion, you can bet it won't be too long before one or more of the
technologies described in this feature is sitting inside devices in
our pockets. Before that happens, though, the runners and riders in
the supermemory steeplechase have just a few hurdles to clear.

The longest-standing pretender to flash's crown is magnetoresistive
random access memory, or MRAM. Under development by several companies
since the 1990s, MRAM chips store information within two thin layers
of magnetic material, each divided into a grid of cells. One layer is
a permanent magnet whose direction of magnetisation does not change.
The other is a temporary magnet whose magnetisation can be flipped 180
degrees by applying a small magnetic field or electrical current. The
relative alignment of the two layers' magnetisations determines
whether a bit is set to 1 or 0 (Science, vol 308, p 508).

MRAM's use of magnetisation is both its strength and its weakness: its
strength because magnetisation is fast and easy to control, allowing
memory to be written and read in as little as a nanosecond; its
weakness because changing the magnetisation of one cell tends to
affect its neighbours too.

This "cross-talk" is a tough nut for MRAM researchers to crack. "They
haven't really been able to solve the problem yet," says James Scott,
a physicist at the University of Cambridge. At the moment it limits
the size of MRAM chips to 32 megabytes, less than one-thousandth of
the capacity of the best flash devices. Electronics companies such as
Hitachi and Toshiba continue to work on improved designs, maintaining
faith in the potential of electrically controlled MRAM for fast,
high-density memory.

o Size: ?
o Speed: ok
o Stability: ok
o Power consumption: ?
Ferroelectric random access memory, or FeRAM, is a close relative of
flash. Like flash, it uses electrical effects to control a
transistor-like structure. But rather than controlling flows of free
electrons, it takes advantage of the strange distribution of electric
charges found in complex crystals known as ferroelectrics.

In a ferroelectric, small external electrical fields can induce
positively and negatively charged ions in the crystal to shift in
position, creating a stable electrical polarisation not unlike the
field between a magnet's north and south poles. Upwards and downwards
polarisations are the 0s and 1s of the ferroelectric bits (Science,
vol 315, p 954). A small voltage applied to the crystal can be used to
send in additional charges, changing the polarisation and causing the
bits to flip. This process is fast - it takes less than a nanosecond
in principle - and requires little power, two of the advantages of

As with MRAM, though, FeRAM's strength is also its Achilles' heel.
"The problem is that FeRAMs are charge-based," says Rainer Waser, a
physicist at RWTH Aachen University in Germany. To switch the
ferroelectric with sufficient speed, the additional charge needs to be
stored somewhere close by, so every FeRAM memory cell comes with a
capacitor attached, eating up valuable space. "The capacitor footprint
limits storage density," admits Scott, who has studied ferroelectric
materials for three decades. "I can't see FeRAMs going to gigabyte
devices like flash."

It could still have its uses, though: FeRAM's low power demands and
straightforward design could make it the memory of choice where
economy is more important than capacity. Toshiba is convinced, and
announced a prototype 128 MB FeRAM chip in February 2009.

o Size: ?
o Speed: ok
o Stability: ok
o Power consumption: ok
When it comes to downsizing to the tiny scales needed to replace
flash, a chameleon technology known as phase-change random access
memory, or PCRAM, looks a promising bet.

It exploits the same sort of technology used in rewritable CDs and
DVDs. These store information in the atomic structure of materials
with two distinct solid phases: an amorphous phase similar to that in
window glass, in which the atoms are arranged in no particular order,
and an ordered, crystalline phase such as that found in metals. The
crystalline state is electrically conducting, and the amorphous state
is an insulator (Nature Materials, vol 6, p 824).

In PCRAM, this material is held between two electrodes. All that is
needed to flip it between its two phases is a pulse of laser light or
electric current applied to the electrodes to melt the material. If
the current pulse is long, the material orders itself into its
crystalline state. If the pulse is short, the material cools abruptly
into the amorphous state (see diagram).

The approach is not without its problems. Heating memory elements to
the few hundred degrees Celsius necessary to change the state
dissipates a lot of power - although that power requirement will sink
as the devices shrink.

With PCRAM there could be a lot of room at the bottom. Only a few
atoms are needed to create a memory unit capable of distinct amorphous
and crystalline states. Luping Shi of the Agency for Science,
Technology and Research (A*STAR) in Singapore reckons that memory-unit
sizes of just 5 nanometres across should be possible - about one-tenth
the size that flash memory has so far attained.

What's more, PCRAM's switching times can be blisteringly fast. "Speeds
of 1 nanosecond are feasible," says Matthias Wuttig of RWTH Aachen.
The problem is that the faster a material is switched, the less stable
its crystalline phase tends to be, so PCRAM speeds are still 10 to 100
times slower than that. With individual bits already being imprinted
on just a few dozen atoms, the challenge now is to work out what
particular combination of different atoms provides the optimal
trade-off between speed and stability. Many companies are working on
that, and Samsung has recently brought out a 512 MB PCRAM memory chip.

o Size: ok
o Speed: ok
o Stability: ok
o Power consumption: ?
PCRAM is not alone in its potential to work at the tiniest of scales.
A rival technology called resistive random access memory, or RRAM,
makes that claim too. Whereas PCRAM relies on heat-induced changes in
a material's atomic structure, RRAM exploits electrochemical reactions
that change the bond structure of certain crystalline solids.

RRAM's raw material is a naturally insulating oxide, such as that of
titanium and oxygen. When a large voltage is applied to such a
crystal, the electron bonds that moor the oxygen atoms to the crystal
start to break. As the oxygen floats off, it leaves behind it both
holes in the crystal and excess electrons that are available for

The holes tend to be aligned in rows, creating extremely narrow,
electrically conducting channels in the crystal. Reverse the voltage
and the oxygen atoms move back towards the channel, cutting electrical
conduction and returning the crystal to an insulating state.

This reversible transition creates stable memory states that only a
high voltage of the right polarity can switch. Once this voltage is
applied, just a few oxygen vacancies moving in and out of the channel
are enough to toggle between conduction on and conduction off, making
RRAM a fast, low-power technology (Nature Materials, vol 6, p 833).

"We can switch our devices in a nanosecond or less, and the energy
required is in the order of a picojoule," says Stan Williams, who
works on RRAM at Hewlett-Packard Laboratories in Palo Alto,
California. That's about one-hundredth of the energy required by
flash. And because the conducting filaments are so small, the
switching process could potentially happen on a scale of just a
nanometre or so, giving RRAM truly tiny potential.
We can switch our devices in a nanosecond, and the energy required is
in the order of a picojoule

Stability is a growing challenge at smaller scales, though. If a
high-resistance bit is set right next to a low-resistance one,
electrical current tends to bypass the high-resistance region and take
an undesirable detour through the neighbouring element. This is a
problem that Hewlett-Packard and other companies are now trying hard
to solve.

RRAM is not just exciting for its conventional memory capabilities. In
2008, Williams and his colleagues realised that RRAM devices have all
the characteristics of a memristor - a fabled fourth basic electronic
element to join the ranks of resistor, capacitor and inductor.
Memristors differ from ordinary resistors in being able to adopt any
number of values for their resistance, according to the current that
flowed through them in the past. That could make them models for the
analogue computational elements inside the human brain - but with a
twist. "These electronic synapses are much smaller and faster than the
synapses in the brain, and use less energy," says Williams.

o Size: ok
o Speed: ok
o Stability: ?
o Power consumption: ok
Racetrack memory
Most routes to supermemories involve finding new ways to manipulate
atoms and their properties on the nanometre scale. Stuart Parkin of
IBM's Almaden Research Center in San Jose, California, believes we
should instead rethink memory design. "Maybe considering entirely new,
three-dimensional architectures will enable us to improve memory
devices further," he says. He and his IBM colleagues have got just
such a suggestion: racetrack memory (Science, vol 320, p 190).

With racetrack memory, bits are stored as tiny domains of opposing
magnetisation, rather as they are in a conventional hard drive. The
difference is that the memory units, or domains, are not carved on a
monolithic block, but strung out like pearls along a nanoscale
magnetic wire. An electric current shunts these domains along the wire
and past special reading and writing heads, where the information
stored in the pattern of bits can be retrieved or modified (see
diagram). This can be done at speeds of up to 200 metres per second,
resulting in read times of tens of nanoseconds.

The big potential benefit of the racetrack is its storage capacity.
Even just a flat micrometre-sized wire could store information with a
density comparable to that of flash, says Parkin. The real deal comes,
though, if the nanowires deviate from a standard two-dimensional
configuration and are instead coiled into a three-dimensional
arrangement of mini-skyscrapers. Then, hundreds of times more bits can
be stored than in flash memory covering the same area.

So far, only two-dimensional prototypes are in development, which can
match the storage density of flash. For 3D racetracks, Parkin admits
his team will need a little longer. If the skyscrapers get off the
ground, though, computer memory might have a very different face
before too long.

o Size: ok
o Speed: ?
o Stability: ok
o Power consumption: ok
Editorial: Writ small, huge recall
Flash: memory hero
Conventional computer hard drives, with their mechanical arms that
read information from spinning magnetic discs, are power-hungry,
comparatively bulky and prone to failure. Flash memory, developed by
researchers at Toshiba in Japan in 1980, is compact and demands little
power. That's why it has rapidly come to dominate the market for
small-scale permanent computer memories, despite its higher price tag.

Flash memory bits work in a similar way to the transistors that
toggle currents in a computer's processor chips. In a transistor, a
tiny electron-conducting channel is topped off by a metallic strip
known as the gate. A voltage applied to the gate creates fields that
determine whether electrons can flow through the transistor channel,
producing controllable on and off states - the binary 1 and 0. In
flash, the only difference is that the electrons are trapped at the
gate by a surrounding layer of a highly insulating oxide, making the
on and off states permanent.

The memory state can only be changed by applying a large voltage that
allows the electrons to escape through the oxide. This makes flash
drives slower than hard drives, and the passage of electrons also
slowly degrades the oxide's insulating capabilities. Flash memory can
be rewritten only so often - generally between 10,000 and 100,000
times - before failing.

Flash's main limitation is storage density, however. The best flash
chips currently have a storage density comparable to that of magnetic
hard drives, but in both cases significant further miniaturisation
will be difficult. For flash, quantum effects such as electron
tunnelling will make the memory patchy if the bit-storage size dips
below about 20 nanometres.

More information about the AccessIndia mailing list