.jpg)
.jpg)
Liquid
Logic
By Mark K. Anderson
Moore's law is a death march. In a decade or two, the silicon chip will be kaput. What then?
The year is 2015. Computers are fast - really fast. But there's a supercharged black box that puts the whole microchip drag race to shame. No one now knows what it'll be called, but this much is certain: The letter Q will be right up front.
Q stands for quantum, and it just may replace e and i as the tech prefix of choice. Don't hold out for a qMac anytime soon, but even in its embryonic state, the quantum computer is already turning heads. The technology is based on two facts of life at the submolecular level. First, quantum particles such as electrons can exist in multiple states at once. Second, particles in a group can become so intertwined that the actions of one affect all others at the same instant, allowing engineers to build circuits out of individual atoms.
Legendary physicist Richard Feynman noted in the early 1980s that, put together, these two traits would open the door to computational power inconceivable on conventional computers. A working QC only a few hundred bits in size could outstrip Moore's humdrum doubling by astronomical amounts for certain applications. But what's not widely known is how many competitors are vying to be the first to produce a new warp-speed machine that leaves behind the pathetic benchmarks of classical 20th-century silicon.
Some, like Bruce Kane at the University of Maryland, are working to push silicon tech down to submicroscopic scale. Others, like Bell Labs veteran Phil Platzman, want to replace solid-state electronics with a new breed of supercooled liquid computer.
Today, the bit is king. A conventional computer is just a series of hundreds of millions of switches whose on and off positions represent the values 1 and 0. And every operation a computer performs, whether calculating your waiter's tip or simulating the explosion of a nuclear warhead, comes down to a series of actions that flick those switches from 0 to 1 and back again.
Familiar stuff. But in a QC, the bit is upgraded to a quantum bit, or qubit, that doesn't need to choose between 1 and 0. It can be both at once. As a result, a memory array of n qubits can represent every number between 1 and 2n simultaneously.
A QC's capacity doubles with each additional qubit. It may be humbling that the world's largest QC is currently only 7 qubits in size, and can barely process single-digit numbers. But a QC of 333 qubits would be able to perform operations instantaneously on every number between 1 and a googol (10100), a value considerably larger than the number of atoms in the universe. To carry out addition or multiplication on every positive integer between 0 and 10100 would take one of today's supercomputers several quadrillion years as it marched through one number at a time. But a QC would perform the calculation all at once, and it'd be done.
The basic technology behind today's 7-qubit prototype at Los Alamos National Laboratory may be familiar to anyone who's ever had an MRI scan. Nuclear magnetic resonance (NMR) works at the subatomic level, where particles are small enough that they answer to the fuzzy laws of quantum mechanics, and bits turn into qubits. The nucleus of an atom is an electrically charged spinning ball, causing it to act like a bar magnet. Each nucleus has magnetic north and south poles that wobble and rotate together, like a buoy in a stormy electromagnetic sea that responds to the waves lapping at its sides. Hit the buoy's resonant frequency and you can flip it over like a kayak, and right it again, too. Within the molecule, each nucleus can be linked to its neighboring nuclei through the quantum behavior called entanglement - an all-for-one, one-for-all state in which one qubit's actions affect all others it touches. So a chain of atoms can be rigged up with the conditional logic - such as AND, OR, XOR - to make a computer.
Computation in an NMR computer is done by beaming in pulses of radio waves tuned to the particular resonant frequency of each nucleus in molecules of a liquid solution, such as chloroform and crotonic acid, and detecting the resonant frequencies emanating from the resulting nuclear alignment. Each nucleus' resonant frequency changes depending on whether its neighbors are in their 1 or 0 states, so radio pulses can be used as sheepdogs to herd the qubits through an algorithm's flowcharts. Unlike a conventional computer, each branch of a flowchart represents not an either-or choice, but rather a bifurcation: One state of the computer answers "yes" and follows the instructions from there, while another simultaneous state answers "no" and does likewise. For the right kind of calculations, such as the factoring of large numbers, these superimposed splits build up an exponential speed advantage over classical bit-based logic. (The biggest challenge for the programmer is choosing which of many simultaneous answers to convert back to 1s and 0s for output.)
Ultimately, however, NMR computing is a dead end. Each qubit and logic gate needs its own signature resonant frequency, and there are other problems involving initialization and readout of the system. "NMR will stop in the next few years in terms of number of qubits," says Los Alamos mathematician Emanuel Knill, who codesigned the 7-qubit machine. "It's like the development of classical computers - at some point we had to switch from vacuum tubes to something else."
Mark K. Anderson (mark@markkanderson.com) has written for Science, Harper's, and Wired News.
