Company says it has a substitute for the transistor that's far smaller and, potentially, far cheaper to produce.
Staff Writer, CNET News.com
With a recent breakthrough in making circuits with molecules, Hewlett-Packard hopes to change chip history and expand its own role in the process.
Researchers from the Palo Alto, Calif.-based computing giant have created devices called crossbar latches that can be used to perform calculations in microprocessors, the same function silicon transistors now have.
The difference is that crossbar latches—which consist of a grid of microscopic wires linked by molecules at their intersections—are far smaller and, potentially, far cheaper to make because they are produced using processes more akin to inkjet printing rather than the ornate etching processes required for today's chips. Both factors give chipmakers an opportunity to dodge some of the technical difficulties and painful costs awaiting them in coming decade.
HP has already shown how crossbar latches can be used in memory.
"This is the final piece of the puzzle for building a molecular computer," said Phil Kuekes, senior computer architect and primary inventor at HP's Quantum Science Research (QSR) unit.
Adoption of crossbars across the industry could also lead to royalties for HP, which may try to license it, added Stan Williams, director of the QSR. HP is so confident of its technology that is aiming to get elements of crossbar technology incorporated into 32-nanometer chips, which will hit commercially in 2011 or 2012. The company will try to get its technology ensconced in industry road maps guiding equipment makers and semiconductor designers.
"There is a recognition that there is going to have to be innovation," Williams said. "We'd like to introduce some aspect of it into that (32-nanometer) node."
HP, however, isn't trying to find a way out of the conflicts many semiconductor designers face. Researchers from the South Korea, Japan and the United States—including IBM and Intel—will next week publish papers at the International Solid-State Circuits Conference detailing ideas for new types of chips and transistors.
"The single biggest advantage we have is that we can do it now," Kuekes said. "We think we can make complex devices sooner."
Although experts and pundits have declared the imminent death of Moore's Law for three decades, the end appears to be in sight. The principle, which states that chipmakers can double the number of transistors on a silicon chip every two years, has enabled the industry to shrink the size and cost of things like computers and cell phones while improving their performance.
Unfortunately, traditional silicon transistors can't be shrunk in size much longer. Circa 2021, there won't be enough atoms inside traditional transistors to contain the flow of electrons. Hybrid chips that contain elements of traditional silicon chips and some undetermined materials or structures will appear in the first half of the next decade, and chips based on the new materials are predicted to emerge in commercial production in the 2020s, if not earlier.
What it looks like
A single crossbar latch consists of a three wires: a "latch" wire and two control, or clock, wires. The latch wire lies under the other two. The wires are connected by molecules, which transfer electrical impulses from one wire to the next. (In the latches used to perform calculations, it is a layer of a common acid made up of carbon, hydrogen and oxygen.)
In layman's terms, a series of electrical impulses will close the molecular switch between the latch wire and the first clock wire. The impulses will then open the switch between the latch wire and other clock wire. In digital terms, a computer interprets this action as a "0". Conversely, opening the first switch and closing the second becomes a "1."
Earlier, Kuekes had produced crossbar latches that could perform basic calculations, but they couldn't store partial results for later usage. The new crossbar latches, however, detailed in an article in the Journal of Applied Physics, can: They conceivably perform transistorlike functions.
A key attribute of the switches is that the junction between the wires can be as small as 2 nanometers. The equivalent junction in current transistors inside 90-nanometer chips is about 60 nanometers, meaning that far more crossbar latches can be put into the same space that now holds transistors. Traditional transistors, in fact, will never be able to hit these limits, Kuekes said.
"The three most important things are size, size and size," he said. "When you get down to around 15 nanometers, the physics of semiconductor transistors will not work."
Shrinking the electrical junctions in a chip also generally increases performance, but the switches in the experimental crossbar latches only flip at about a tenth of a second.
Just as important, chips made on crossbar latches could be cheap to manufacture. The wires are put into place through nano-imprint lithography. In this technique, a customized mold is placed into a film later; the imprints left by the mold become the templates for the wires.
The molecular switches, meanwhile, do not have to be placed individually at the juncture of the wires. Only wires at the junctions will carry a current.
"Essentially, all of the other molecules are sacrificed," Williams said.