Manufacturing Ultimately Drives Moore's First Law

The follow up to this conclusion is that as you continue to shrink the process—getting fine and finer line tracings and so on—the equipment need to control this and make it happen needs to get more and more precise. Which in turn means that each generation of manufacturing machines will cost more than the prior one.

For example, at HP we've shifted from using light to ultraviolet light to try to get better resolution on making the chip traces. This is because ultraviolet light has a shorter wavelength, allowing you to more accurately measure and produce these traces. Whenever you make a major shift to change and improve the line widths that you can produce on the chip, you basically have to build an entire new factory.

In fact, chipmakers must continually invest in new technologies. Consider that in 2000, chipmakers were building circuits 180 nanometers wide. To put that in perspective, you could line up 500 of these circuits and they would just fit across a human hair. By 2010, the industry standard is predicted to be an even more amazing 45 nanometers wide.

The entire field of semiconductor manufacturers is running into a barrier that exists in no other technological field, with the possible exception of those designing electron microscopes: the limits of optical lithography. In the normal manufacturing process, light beams are used to trace out the patterns of circuits. However, it may well be that soon even ultraviolet wavelengths are reaching a point where it's difficult to build lenses to work with the equipment effectively.

Gordon Moore postulates that eventually as manufacturers move away from optical lithography, the subsequent technique to employ in order to continue making progress that sustains the first law is the use of even shorter wavelength X-rays.

Most of the time, those outside of the semiconductor industry are amazed to hear this. True, chip making is at its base a manufacturing process, albeit a highly technical one. However, it's much different than an automobile assembly line. On an auto production line, the tolerances between designs—how well pieces fit together—are usually not measured in microns.

Shifting a plant from using Type A door to Type B hatch is normally not much of an issue. That being the case, plants can shift from building lower selling types of compact cars and change over to manufacture higher volume sport-utility vehicles in a matter of days or even hours. Contrast that with the high R&D costs for testing and building all new equipment to measure out and build chips that may be trying to cram more than double the number of transistors on a microprocessor. You start to see why changeover isn't quite as easy at it would be for a number of industrial processes.