|
Thursday, June 9, 2005
HP Proposes Error Correction Scheme for Nano-electronic Circuits
|
|
You are sending an email that contains the article
and a private message for your recipient(s). |
Your Name: |
|
Your e-mail: |
* Required! |
Recipient (e-mail): |
* |
Subject: |
* |
Introductory Message: |
|
HTML/Text
(Photo: Yes/No) |
(At the moment, only Text is allowed...)
|
|
|
Message Text: |
HP today announced that its researchers have created a new way to design future
nano-electronic circuits using coding theory. The result could be nearly perfect
manufacturing yields with equipment a thousand times less expensive than what might be
required using future versions of current technologies.
In a paper appearing in the June 6 issue of "Nanotechnology," a publication of the
Institute of Physics, HP Labs authors Phil Kuekes, Warren Robinett, Gadiel Seroussi and
Stan Williams explain in detail a defect-tolerant interface to HP's patented crossbar
architecture.
"We have invented a completely new way of designing an electronic interconnect for
nano-scale circuits using coding theory, which is commonly used in today's digital cell
phone systems and in deep-space probes," said Williams, HP Senior Fellow and director,
Quantum Science Research at HP Labs. "By using a cross-bar architecture and adding 50
percent more wires as an 'insurance policy,' we believe it will be possible to fabricate
nano-electronic circuits with nearly perfect yields even though the probability of broken
components will be high."
Williams said he believes future chips will have to rely, at least in part, on the
crossbar architecture, in which a set of parallel nanoscale wires are laid atop another
set of parallel wires at approximately a 90 degree angle, sandwiching a layer of
electrically switchable material in between. Where the material becomes trapped between
the crossing wires, they can form a switch that represents a "1" or "0," the basic
building blocks of computer code.
Future chips may be limited in the geometric complexity that can be created at the nano
level because of problems with precision alignment. Crossbar structures are highly
regular and therefore relatively easier and less expensive to fabricate than the complex
array of wires, transistors and other elements in today's processors. The disadvantage of
crossbars is that they require more space on the silicon substrate.
"We think the tradeoff of space versus manufacturing ease will become more an issue in
the near future," said Williams.
Furthermore, as sizes of electronic features get down to a few nanometers, it will become
either physically impossible -- or economically unfeasible -- to produce absolutely
perfect circuits.
"Future chip manufacturers will have to deal with the reality of defects," said Williams.
HP's approach involves enhancing a device known as a demultiplexer, which enables data to
be read and written in a circuit by connecting the crossbar array of nanowires to a small
number of conventional wires. By adding a few more conventional wires and using basic
coding theory, the HP researchers show that the demultiplexer will still work even if a
significant number of the connections between the conventional wires and the nanowires
are broken.
"It's like giving a distinctive name to a restaurant host to be sure you hear your party
called above the noise of the crowd," said Kuekes, a senior computer architect and one of
the authors of the paper. "Instead of 'the Jones party,' you might put yourself down as
'the John Paul Jones party.' That way, when the host calls your name, you'll hear it,
even if every word doesn't come through clearly."
Using defect tolerance to replace the need to produce "perfect" chips could provide a
huge cost advantage for chip manufacturers in the future.
Williams said the HP Labs group has created working devices in the laboratory at the 30
nanometer half-pitch scale -- about a third the size of today's chips. The International
Technology Roadmap for Silicon, the standard for the industry, predicts that chips using
features at 32 nanometers half pitch should be in production in about seven to eight
years. |
|
|
|
|