Sometime in the next 15 years, the laws of physics decree that the transistor will finally stop shrinking, and when it does, engineers will need an alternative. Recent advances seem to improve the odds for a possible successor, a nanometer-size switch under development at Hewlett-Packard Co., in Palo Alto, Calif.
HP has been studying the switch for some years and has already built it into experimental memory and logic circuits. But only now have its researchers found a way to give the switch "gain"--the power to amplify a signal. This power is one of the great features of transistors, because it allows a signal to pass through a circuit without petering out.
"That's what makes this development interesting," says Philip J. Kuekes, senior computer architect at HP Labs. "With power gain, we can build things of arbitrary complexity without having to cheat and go back to silicon transistors."
The switch, a tunnel junction, consists of a single molecular layer sandwiched between two metal wires. The layer is normally an insulator. But if it is thin enough, charged particles can pass through-a quantum-mechanical phenomenon known as tunneling. If the junction's resistance to such tunneling current is high, the switch is open; if it is low, the switch is closed. Applying a high or low voltage across the switch toggles the barrier between low and high resistance.
HP achieved the gain it needed by using what it terms a "molecular crossbar," made from two switches. Each switch connects to its own "control" wire, which supplies it with voltage pulses. The two switches also connect to a common latch wire, which carries either a high or a low voltage. Together the two switches encode a single bit.
Logic Chopper
In HP's crossbar latch, two very thin switches cross a latch wire [micrograph]. Each switch is opened or closed by a voltage pulse from its own control wire and the voltage level of the shared latch wire. Together the two switches store a single bit--either a 1 or a 0, depending on which switch is open and which is closed.
To enter information, the researchers apply a series of pulses through the control wires. The first pulse opens both switches. Then a second pulse is applied to just one of the switches, closing it if the voltage on the latch wire is high (conveying from a previous circuit the value 1). A third pulse is applied to the other switch. If the latch voltage is high, that switch stays open. If the voltage is low, that switch closes [see figure].
The combined state of the two switches--that is, which one of them is open and which one closed--stores either a logical 1 or a logical 0--one bit.
In a real system, the latch wires would connect a series of logic circuits. Normally, the voltage on the latch wire would degrade as the signal passed from one circuit to the next. The crossbar latch solves this problem, though, by bringing the voltage back to the right level, either high or low.
Here is how it works: one switch receives the high voltage level, and the other gets the low voltage--one close to ground. Only the switch that is closed will let the voltage through to the latch wire; once there, it restores the wire's voltage to the high or low level. The result is gain.
Kuekes, R. Stanley Williams, and their colleagues at HP realized that the crossbar system had another advantage: it could make possible additional logical function. Besides "AND" and "OR," it can also perform "NOT," which changes the latch voltage from one level to the other. With gain and a full palette of logical functions, the HP nanoswitch can exploit its greatest advantage over the transistor: its shrinkability.
The critical dimension is the thickness of the barrier inside the tunnel junction, just 2.8 nanometers, the diameter of a single molecule of stearic acid. HP says other materials could also be used. Other parts of the circuitry are still much larger, in the micrometer range.
Now the researchers plan to build a molecular device whose every dimension lies in the nanoscale and to knit billions of them into a huge circuit. It won't be easy. They're talking of getting the technology on the market in about 10 years.