Portland, Ore. Hewlett-Packard Co.'s HP Labs has designed a crossbar switch that may one day allow the company to pack the power of a traditional microprocessor arithmetic logic unit into just a few square microns.
Current bus structures applied at the nanoscale are akin to trying to drain a swamp with a soda straw domains measuring a few nanometers across are just too much of a bottleneck for traditional input/output schemes.
Hewlett-Packard (Palo Alto, Calif.) claims that its crossbar switch design has broken that bottleneck by reducing the I/O problem by 128 to 1, while providing a bistable latching architecture for cascading any number of layers of nanoscale memory and logic.
"A single pair of lines can now control 100 or more switches at the same time," said Phil Kuekes, senior architect at HP Labs' Quantum Science Research group. "Since our latches can also restore the logic levels of these bits, we can now gang together any number of layers of logic and memory functions. Next we want to show we can perform traditional logic functions with crossbar switches, but at a density you can't achieve today with transistors such as packing a 128-bit ALU into just a few square microns."
In 2002, HP Labs demonstrated its first two-terminal platinum/organic-monolayer/titanium nanowire array of crossbar switches. They were fabricated with optical lithography from a bottom electrode made from 100-nanometer-thick platinum (Pt), a Langmuir-Blodgett atomically thin monolayer of organic rotaxane and a top electrode of titanium. The array was built atop a polished silicon wafer capped with 200 nm of insulating silicon dioxide.
HP Fellow Stanley Williams, director of the Quantum Science Research group, shared the patent on that crossbar switch design with HP Fellow Kuekes and University of California-Los Angeles scientist James Heath. That crossbar grid of nanoscale wires formed an array of crossbar switches. The crosspoint of any one of those switches could be selected by virtue of electrically activating the rotaxane molecules between them. "What we had before was just a memory we always knew whether we were writing a 1 or a 0. But now what we write to memory is sensitive to the signal line value," said Kuekes.
Essentially, he said, "We have replicated the important properties that digital circuits use transistors for, but with much smaller, nanoscale crossbar architectures. We are not getting gain, so we cannot perform any of the analog operations that transistors perform, but we have been able to demonstrate both memory and logic functions without transistors."
In the new design, HP substituted an organic cadmium stearate salt monolayer 2.8 nm thick for the rotaxane. The top electrode, of 5-nm-thick titanium plus 200 nm of aluminum, was evaporated through a shadow mask. The nanowire crosspoints had a lateral nanowire width ranging from 1 to 10 microns (Pt) and 5 to 20 µm (Ti/Al), yielding an active junction area of 5 to 200 µm2. The monolayer itself had less than a billion molecules electrically in parallel to enable switching between a measured 3,000 ohms when closed to over 10 megohms when open.
In terms of speed and longevity, the performance of this prototype chip was dismal. HP argues, however, that the logic and computing functions it made possible were the important breakthroughs. It will be easy, said the company, to speed the crossbar architecture and raise its mean time to failure with further device characterization.
Although "the key here was not performance," Kuekes said HP is "trying all types of different organic molecules to increase speed and longevity. The important thing here is that we can now get one switch to open and another to close or the first to close and the second to open based solely on the logic value on a common signal line. Since all the molecular junction sees is the total voltage across it, in our architecture we arranged it so that the value of the signal line determines whether each bit switches or not."
In this novel approach, instead of setting or resetting a grid line by sending a fully-on control pulse to set the bit, the experimenter can send a smaller, conditional pulse to both the set and reset control lines simultaneously. Only the one whose logic states matches switches on.
HP researcher Duncan Stewart joined the original team for the new design.