New Approaches to Computing

There are a lot of radical ideas out there about how to make calculation much faster than it is today. We'll just mention a few.

Supercooling

Electrical circuits operate faster and more efficiently when they are cold—very cold, like 90° Kelvin (around −300° Fahrenheit). The biggest benefit of the CPU-in-the-deep-freeze is that it makes much higher clock speeds available. Unfortunately, there are a lot of hurdles to overcome before your desktop will feature a really cool computer. One is that the technology currently needed to supercool electronics is both expensive and bulky. If that problem is solved, there is the fact that both the CPU and the circuitry that it connects to will have to be redesigned to operate successfully in this temperature range. Still, these are not huge barriers. Supercooling is already being used in some high-end computers (supercomputer class), and a variety of companies are working to bring it to the midrange of mainframe competitors.

On a slightly different front, engineers are exploring the use of materials in the chip itself that have the property of carrying heat away. While this approach won't create dramatic changes in speed, the expected 100 percent or so improvement would be significant.

Optical Computing

As will be mentioned frequently in this book, electrical signals are vulnerable to interference from many sources. Reducing this vulnerability to zero is probably impossible; even getting close would require an enormous amount of bulky shielding. The whole concept of "micro" would disappear. Computers compensate for the probable interference by using complicated algorithms that analyze information sent from "A" to "B" in order to determine if it arrived correctly. This is true even within the CPU; it becomes a major exercise as data flows off the CPU to other parts of the computer.

An optical computer would be a great solution to this problem. Light in a vacuum is faster than electricity in a wire (∼30 cm/sec. vacuum vs. ∼20 cm./sec), but with current technology, this difference is of relatively little value. More important for the moment is that it is much easier to guard against interference in optics, which means that data carried by light can be transmitted with less error checking. The throughput (the real vs. the theoretical speed) of optical switches is therefore much greater than with data carried by electrons. Research has demonstrated that internal connections in a computer, say from the CPU to memory or to disk, could be accomplished much faster by light than by electricity. And it is feasible to build such a connection. So why isn't it done?

The first reason is that there isn't yet a compelling need for such speed in mainstream commercial machines. Existing computers are not yet hobbled by the problem of moving data on their very short, relatively error free, internal paths. This will likely change once the annual sixty percent or so increase in CPU speeds starts to go away—perhaps around 2015. The more important reason why optical components are not yet used inside production systems is that the increases they would bring are, at the moment, largely theoretical. The problem is that only an all-optical device will be significantly faster—translating from electronic data to optical data is a serious bottleneck. The electronics that do this are complicated and not especially fast.

So why not make the whole thing, including the CPU, optical? That's a powerfully attractive idea, and lots of scientists are working on it. Optical computers exist in the lab, but making a purely optical system, one that is able to avoid all the problems of electricity, is not yet possible on a commercial scale. Indeed, researchers appear to be quite some distance from achieving that. They need, as they say, a few "breakthroughs." Of course, breakthroughs are things that may or may not occur. Experience says that they will in this case, but when—two years, ten years, twenty years—is anybody's guess.

In the meantime, computer companies are getting lots of experience with optics. Because interference, and therefore reduced throughput, has been a real problem in computer networking, optical interconnects (fiber optics) are in wide use there and are rapidly becoming ubiquitous.

Even More Exotic Stuff

The electronic logic gate has a long way to go—to around 2010–2015—before it begins to lose momentum as the center of computation. Even so, scientists are actively exploring other, radically different techniques. One approach is to directly manipulate atoms so that the basic unit of measure becomes individual electrons (quantum computing). Expect to hear a lot about carbon nanontubes in the near future. Another concept is to use biochemical reactions, as in combining DNA, to execute enormously complex calculations at speeds that electronic systems could not hope to duplicate. Indeed, scientists are exploring pretty much everything that mirrors the ability of a transistor to act as a binary switch. As interesting as these ideas are, they aren't likely to have much of an impact in the near future—most definitely including the useful lifetime of this edition.