SAN FRANCISCO Work at IBM Corp. on the theory and practice of quantum computing suggests that the industry may be closer to practical CPUs that could process information in the form of quantum bits, or "qubits," rather than conventional binary bits.
The new thinking was discussed today (Dec. 11) in a plenary lecture at the IEEE International Electron Devices Meeting here. David DiVincenzo of IBM's T. J. Watson Research Center (Yorktown Heights, N.Y.) surveyed the prospects for quantum computing, concluding that practical, solid-state devices may soon emerge to support the theoretical projections of vast computing power arising from this technology.
"The principles and promise of quantum computers," DiVincenzo said, lie in the "requirements for the physical implementation of quantum computers in atomic physics, quantum optics, nuclear and electron magnetic-resonance spectroscopes, superconducting electronics and quantum-dot physics."
Proving a point
DiVincenzo already has a track record in quantum theory. For instance, in 1995 he mathematically proved that two-qubit operations were sufficient to execute any quantum algorithm. Thus, engineers need not design more than two-qubit physical devices to reap all the parallel-processing benefits of any future quantum algorithm.
In DiVincenzo's view, all quantum devices will require a new formal basis to express the kind of algorithmic parallelisms that could be realized with quantum computers. Unlike conventional encodings of information, quantum devices allow the superposition of multiple discrete states simultaneously on the same qubit. Thus, multiplying two qubits together is equivalent to simultaneously multiplying every possible string of values that a conventional computer register could hold.
That kind of operation demands a new mathematical formalism in order to craft effective quantum algorithms, said DiVincenzo. "Making bits that obey the quantum-mechanical principles [and] efficient algorithms for some otherwise intractable problems, like prime factoring, becomes possible," he said.
DiVincenzo listed a sevenfold set of requirements for physical implementation of quantum computing. They are scalable well-defined qubits, resettable states, long superposition times, a universal set of quantum gates, easy qubit measurements, easy qubit-to-digital conversions and easy qubit telecommunications.
Today, laboratory implementations of quantum devices concentrate on perhaps one or two of the seven requirements, he said, but to create commercial devices all seven must be met.
DiVincenzo summarized current attempts at building a quantum "transistor," most of which only have a superficial resemblance to today's silicon transistors. Using a voltage-controlled gate to switch the qubit may be the only recognizable commonfeature.
The most striking new feature, according to DiVincenzo, is the fact that a single atomic element will embody the memory element in a microscopic domain, probably in the spin direction (either up or down) of a single atom or electron. The quantum gate will most likely be switched with a voltage-controlled pulse at the gate, which superimposes a new state into the currently executing quantum algorithm.
Aside from these generalizations, there is little similarity among the various current proposals for implementing quantum computers. Methodologies, so far, are based on basic physics rather than chip technologies, DiVicenzo said, but advances in quantum dots embrace standard solid-state physics for integrated circuits. In particular, DiVincenzo described an ion-trap computer that holds qubits in pairs of energy levels of ions held in a linear electromagnetic trap.
Other atomic-physics-based proposals use the position of atoms in a trap or lattice or the vibrational quanta of trapped electrons, ions or atoms as their qubits. The presence or absence of a photon in an optical cavity has also been proposed as the basic qubit mechanism, and superconducting devices are being proposed that store qubits as charge or "flux."
On the solid-state chip side, impurities can introduce well-characterized discrete energy-level spectra directly onto silicon chips. Separately, researchers are at work on various quantum-dot approaches, storing the qubit in the spin state, the orbital state or the charge state of quantum dots, according to DiVincenzo.
"Every implementation detail of a qubit from its initialization to its interaction with neighboring bits, the errors that it might be subjected to and its readout have to be thought out and investigated from scratch," he said.
DiVincenzo has narrowed the overall development effort, however, by applying his seven criteria to the known theories today. The result is to single out, by elimination, a model for the solid-state implementations of the future.
The right spin
In DiVincenzo's model, qubits are represented by the spins of individual electrons trapped in an array of quantum dots. Many proposals for building quantum-dot arrays are on record, but DiVincenzo's model will reset the system by cooling the device to a predetermined temperature. Basic two-bit gates would be built by changing the height of an electrostatic barrier between quantum dots.
According to Divincenzo, the superposition time of such arrays has already been determined in preliminary laboratory experiments measuring spins in semiconductors, to be long enough to allow quantum algorithms to execute. And other laboratory measurements, he said, also suggest that spin can be conveniently converted into directly measurable electron position.