Researchers Find that the ‘Crossbar Latch’ Can Replace Transistors and Improve Processing Power

The often-quoted Moore's Law says that we can expect processors to double in capacity every 18 months. And that has held true. Now computer scientists warn that - without some significant breakthrough - further advances in computing power and speed will be halted in the next decade by the physical limitations of conventional silicon technology.

But scientists in HP Labs have demonstrated that it could be possible to continue advancing computer power by eliminating the need for transistors - the basic building block of computing for the last 50 years. In a paper published today in the Journal of Applied Physics, three members of HP Labs' Quantum Science Research (QSR) group in Palo Alto offer a feasibility-level description and demonstration of the "crossbar latch" - a bistable-switch latch that promises to replace the traditional transistor and to resolve some issues that have stymied progress in molecular scale computing.

While a patent for the crossbar latch was granted in 2003, this article reports on experiments that demonstrate that the theoretical construct actually works. The invention is a key step in going beyond the material limits of conventional silicon processes. Using electronic devices created by trapping an electrically switchable layer only a few atoms thick between crossed wires, a bit of memory can be stored or a logic function can be performed at each intersection of wires.

Standard semiconductor circuits require three-terminal transistors to perform the 'NOT operation' and restore signals. However, it is generally believed transistors will not function at sizes of a few nanometers, and that is why there is a practical limit to their miniaturization.

HP researchers and their academic colleagues are believed to be the first to propose an architecture for molecular electronics based on programmable crossbar circuit arrays populated with two-terminal devices at each crosspoint. At issue, however, has been how to manage signal restoration and inversion in a general computing scheme that does not rely on transistors. The crossbar latch is the solution to both problems.

"We are reinventing the computer at the molecular scale," said Stan Williams, HP Senior Fellow and QSR director, and one of the authors of the paper. "The crossbar latch provides a key element needed for building a computer using nanometer-sized devices that are relatively inexpensive and easy to build." A nanometer is one billionth of a meter or 39/1,000,000,000 inch.

In the journal article, HP Lab researchers describe and demonstrate a microscopic device consisting of a single wire acting as a signal line, crossed by two control lines with an electrically switchable molecular-scale junction where they intersect. By applying a sequence of voltage impulses to the control lines and using switches of opposite polarities, the latch can perform the NOT function essential for general computing operations. In addition, it can restore a logic level in a circuit to its ideal voltage value, allowing a designer to chain many simple gates together to perform an arbitrary computation.

In principle, these operations enable universal computing for crossbar circuits, and - potentially - integrated nanoscale electronics. And that means Moore’s Law may hold on for another 50 years with the potential for processors that are thousands of times more powerful than today.

"We have previously demonstrated that we could make a working memory with molecular-scale junctions and logic devices that could perform simple logic operations such as AND and OR," said Duncan Stewart, QSR scientist and one of the authors of the paper. "With the crossbar latch, we now have the final component theoretically needed for performing the multiple processing steps required for useful computing at the nanoscale."

As Moore's Law begins to collide with the laws of physics, HP Labs (and collaborators at Imaging and Personal Systems Group in Corvallis) is leading an effort to provide a scientifically sound, practical, economical alternative to silicon technology.

"The technical problem we face is to make extremely small electronic components (at the molecular scale), use very large numbers of these components to make very complex circuits, but manufacture these circuits at much less cost than today's integrated circuits," says Phil Kuekes, another author and senior architect, QSR. "This requires many disciplines to work together. To make them function at the smallest scales, we use quantum physics. To achieve massive complexity, we use computer architecture. To keep them inexpensive, we use low mechanical precision and do self-assembly with chemistry. And to deal with the inevitable defects of chemical self-assembly, we use defect tolerant design algorithms." These algorithms enable computer architects to design around flaws in chips.

None of this is going to happen tomorrow. But the way has been paved.

"Transistors will continue to be used for years to come with conventional silicon circuits," said Kuekes, "but this could very well replace transistors in computers someday, just as transistors replaced vacuum tubes and vacuum tubes replaced electromagnetic relays before them."

Kuekes was previously awarded a patent on the crossbar latch (U.S. 6, 586,965) in July 2003, and the Journal of Applied Physics report, entitled, "The crossbar latch: Logic value storage, restoration and inversion in crossbar circuits," demonstrates the application of the technology. Stewart performed most of the testing that demonstrated the device actually works. The paper underwent rigorous peer review before being published.

While a patent for the crossbar latch was granted in 2003, this article reports on experiments that demonstrate that the theoretical construct actually works. The invention is a key step in going beyond the material limits of conventional silicon processes. Using electronic devices created by trapping an electrically switchable layer only a few atoms thick between crossed wires, a bit of memory can be stored or a logic function can be performed at each intersection of wires.

Posted 1 February 2005.

Tell Us What You Think

Do you have a review, update or anything you would like to add to this article?

Leave your feedback
Submit