HP
maps molecular memory
By
Eric Smalley,
Technology Research News
Building electronic components like computer
memory out of individual molecules would yield extraordinarily powerful
and cheap computers. But figuring out how to mass-produce the devices
is a tremendous challenge.
Assuming the devices can be built, another monumental challenge remains:
how do you talk to them? The wires in today's semiconductor
devices are about 100 times too large to fit molecular devices.
Researchers at Hewlett-Packard Company have found a random chemical process
that bridges the gap.
The researchers' proposed molecular memory unit is a grid of tiny wires,
each about two nanometers in diameter. A nanometer, which is one millionth
of a millimeter, is about 10 carbon atoms long. A single molecule at each
junction of the nanowires is an electrically activated switch whose on
and off states represent the ones and zeros of computing. On one side
the tiny wires extend past the grid.
To connect the memory unit to the outside world, the researchers plan
to randomly sprinkle nanometer-size gold particles on the sections of
the nanowires that extend past the grid and then lay down a set of larger
wires on the gold particles at right angles to the nanowires. This second
set of wires, each about 200 nanometers in diameter, is large enough to
make a connection to the macroscopic world.
By using the right concentration of gold particles, the researchers can
ensure that half of the junctions between the larger wires and nanowires
hold individual particles. "There's a purely random, 50-50 chance that
a nanowire is connected to a big wire by a dot," said Philip Kuekes, a
computer architect and senior scientist at HP Labs.
Some of the junctions the larger wires make with a single nanowire will
have connections and others won't. For instance, a nanowire connected
randomly to 10 larger wires might have connections at the first, second,
fourth, seventh and ninth, but not the third, fifth, sixth, eighth and
tenth larger wires. If a connection represents a one and no connection
a zero, this particular string of junctions would represent the binary
number 1101001010. "So there's a random binary number. That's a unique
address for the nanowire," said Kuekes.
In order to read or write to a memory array of nanowire junctions, you
have to be able to identify each junction, which holds one bit of data.
The binary numbers of the two nanowires that intersect at a junction combine
to make a unique address for the junction.
If it were possible to assign addresses directly to the nanowires, 10
larger wires would be sufficient to name 1,000 nanowires because 210
is 1,024. But because the addresses are assigned randomly, many of them
are duplicated. Increasing the size of the addresses by adding more larger
wires reduces the number of duplicated addresses, said Kuekes.
The trick is finding the balance between getting as few duplicated addresses
as possible and keeping the number of larger wires manageable. The HP
researchers found that four times the log of the number of nanowires is
optimal, said Kuekes. The log of a number is how many times you have to
multiply 10 to get the number. For example, the log of 1,000 is three
because 103 equals 1,000. By this formula, 12 larger wires
can address 1,000 nanowires, 16 can address 50,000 nanowires, 23 can address
500,000 nanowires, 24 can address a million nanowires and 36 can address
a billion nanowires.
To find all the unique nanowire addresses, the HP researchers came up
with a computer algorithm that measures electrical resistance as the larger
wires are switched on and off. Because each nanowire crosses a unique
sequence of larger wires, it has a unique electrical signature. The process
essentially builds a map of the nanowire grid, said Kuekes.
Figuring out how to exchange information between molecular scale devices
and conventional electronic devices is perhaps the most fundamental molecular
electronics problem, said Tom Jackson, a professor of electrical engineering
at Pennsylvania State University. "The HP [proposal] points in that direction,"
he said. "It's significant [but] there are limitations to it."
One problem is that simply connecting the nanowires to the larger wires
with gold nanoparticles would yield fixed connections that could not be
turned on and off, making it impossible to electrically identify each
nanowire, Jackson said. To get around this problem, the HP proposal calls
for adding a molecular switch similar to those in the memory unit to each
of the nanowire-larger wire junctions linked by a gold nanoparticle.
Putting a molecular switch on each nanoparticle and then forming connections
between the nanowires and larger wires without crushing the molecular
switches is a major but not insurmountable challenge, Jackson said.
Researchers at Hewlett-Packard and the University of California at Los
Angeles are beginning a four-year project to build a 16 kilobit memory
device using the molecular memory technology, said Kuekes.
The researchers' ultimate goal is to pack 100 gigabits, or 100 billion
bits, into one square centimeter of chip space using the molecular memory
technology, he said. That's at least 1,000 times more than is possible
using standard semiconductor technology, he said.
The molecular memory addressing system could be used in practical devices
in five to ten years, according to Kuekes. Beginning in five years the
technology could be used in niche products that require very low-power,
very high-density memory, he said. The molecular memory technology should
match the data capacity of standard semiconductor memory in nine or 10
years, he added.
Kuekes' research colleague was Stan Williams of Hewlett-Packard. They
received a United States patent for their research on July 3, 2001. The
research was funded by the Defense Advanced Research Projects Agency (DARPA)
and Hewlett-Packard. Their ongoing work is also funded by DARPA and HP.
Timeline: 5-10 years
Funding: Government, Corporate
TRN Categories: Biological, Chemical, DNA and Molecular
Computing; Integrated Circuits
Story Type: News
Related Elements: United States patent, "Demultiplexer
for a molecular wire crossbar network," number 6,256,767, July 3, 2001
Advertisements:
|
July
18, 2001
Page
One
HP maps molecular memory
Cartoons loosen
up computer interfaces
Virtual view helps
run tiny factory
Bioengineers
aim to harness bacterial motion
Lasers spin electrons
into motion
News:
Research News Roundup
Research Watch blog
Features:
View from the High Ground Q&A
How It Works
RSS Feeds:
News | Blog
| Books
Ad links:
Buy an ad link
Advertisements:
|
|
|
|