Engineering memristor:

Control over device could pave way for computers that learn

Any learning a computer displays today is the result of software," says Yang. "What we're talking about is the computer itself – the hardware – being able to learn

By Jamie Beckett, June 2008

 

HP Labs scientists who in April proved the existence of the memristor have made another significant advance toward developing a new type of computer memory that's many times faster than Flash and could lead to analog computers that process information in a manner similar to the human brain.

The scientists have now successfully engineered control over how the device functions. This means it is now possible to design memristors into integrated circuits that remember information, consume far less power than existing devices and may someday learn from past behavior.

"With engineering control, we can build a device that delivers a specific electrical performance," says Duncan Stewart, principal investigator. "Only then do you get to a point where you can build large integrated circuits."

World's smallest memristive switch

The researchers, members of the Information and Quantum Systems Lab led by HP Senior Fellow R. Stanley Williams, published their experimental findings in the advance online editon of the July issue of the journal Nature Nanotechnology. An earlier paper in the journal Nature (1 May 2008) described the theory of the memristor. 

The team conducted its experiments by building a nanoscale memristor switch – at 50 nanometers by 50 nanometers, it is the world's smallest – that contained a layer of titanium dioxide (a chemical commonly used in both sunscreen and white paint) between two nanowires. As its name implies, titanium dioxide typically comprises one titanium atom for every two oxygen atoms.

Scientist Jianhua Yang found that by subtly manipulating the distribution of the oxygen atoms in this layer, he could control how the device functioned. Although other labs have demonstrated switching using similar materials, none have achieved this level of control over the switches.

 

Controlling the switch

The HP Labs scientists were able to determine both when current flowed through the switch and also how much current flowed through it, operating the switch more like a dial. They could set the switch to 'on' or 'off' – '1' or '0' – and they could dial up or down to anything in between.

"A conventional device has just 0 and 1 – two states – this can be 0.2 or 0.5 or 0.9," says Yang.

That in-between quality is what gives the memristor its potential for brain-like information processing.

 

Faster, cheaper nonvolatile RAM

A memristive device can operate in both digital and analog modes, each of which has different applications.

In digital mode, it could replace today's solid-state memories (Flash) with much faster and less expensive nonvolatile random access memory (NVRAM). That would enable digital cameras without a delay between photos, for example, or computers that save power by turning off when not needed and then turning back on instantly when needed.
 
 Because it is built at nanoscale, the NVRAM chip would also be denser, giving chipmakers the ability to pack more information into a smaller space.

 

Computers that learn

Longer term, in its analog mode, the memristor could possibly enable computers that "learn" what you want.

"Any learning a computer displays today is the result of software," says Yang. "What we're talking about is the computer itself – the hardware – being able to learn."
 
That's not to say the computer would function like a human brain. But it could gain pattern-matching abilities would let it adapt its user interface based on how you use it. These same abilities make it ideal for such artificial intelligence applications as recognizing faces or understanding speech.

 "When John Von Neumann first proposed computing machines 60 years ago, he proposed they function the way the brain does," says Stewart. "That would have meant analog parallel computing, but it was impossible to build at that time. Instead, we got digital serial computers."

Now it may be possible to build large-scale analog parallel computing machines, he says.
"The advantage of those machines is that intrinsically they can do more learning."

 

Proving the memristor

In their first memristor paper published in May, the researchers solved a decades-old mystery by proving the existence of a fourth basic element in integrated circuits. Professor Leon Chua of the University of California Berkeley proposed in 1971 that the memristor should be included along with the resistor, capacitor and inductor as the fourth fundamental passive circuit element.

Although researchers had observed instances of memristance for more than 50 years, proof of its existence remained elusive – in part because memristance is much more noticeable in nanoscale devices.

To learn more, see "Memristor Mystery Solved" or read the paper (subscription required).

The latest paper (authors Williams, Stewart, Yang, Matthew D. Pickett, Xuema Li and Douglas A.A. Ohlberg) is only the latest in a long history of the lab's achievements in the field of nanotechnology. (See below).

 

The lab's many firsts include:

  • Molecular-based logic gate (1999)
  • Crossbar architecture in which parallel wires are crossed by a second set of wires to create a nanoscale electronic switch, delivering memory, logic and integrated memory and logic functions. (1999-present)
  • Highest-density (64 bit) electronically addressable memory known to date (2002)
  • Memory and logic combined using rewritable, non-volatile molecular switch devices (2002)
  • Nano-imprint lithography, a fabrication system that allows an entire wafer of circuits to be stamped out quickly and inexpensively from a master (2002)
  • Crossbar latch (2003)
  • Self-assembling silicon nano-bridge (2004)
  • Nano-crosspoint latch (2005)
  • Field programmable nanowire interconnect design that dramatically improve the density and defect- tolerance of programmable silicon circuits (2007)
  • Memristor existence proven (2008)