In 1971 , Intel astounded the world with its 4004 microprocessor, whose 2300 transistors could execute 60 000 instructions per second. Today, the 820 million transistors of an Intel Core 2 Extreme chip can process nearly 72 billion instructions per second.
Such an improvement is the inevitable result of several decades of Moore’s Law, which refers to the semiconductor industry’s ability to double, every 18 to 24 months, the number of transistors on an integrated circuit. But the chips haven’t seen a commensurate six-orders-of-magnitude cost increase, and that’s because chip manufacturers have had to make those transistors not only smaller but cheaper. In 1963, a transistor cost US $10. That transistor corresponded to half a storage bit and cost as much as an automobile tire at the time. Today flash memory costs $25 for 8 gigabytes, or 64 x 230 bits—enough storage to encode the text of all the books in a small-town library, or more than a 100â¿¿word-per-minute typist could type in his lifetime. And it will be cheaper still by the time you read this article.
If we can keep the pace for several decades more, we’ll see remarkable things: trillion-transistor supercomputers that can track the twists and turns of the world economy; climate modeling that can reliably predict when you should unfurl your umbrella; robots that can mimic human behavior and emotions convincingly enough to make good companions. You’ll be hard-pressed to find anything that can’t use a microprocessor by 2030.
As it turns out, however, that’s a rather big ”if.” For its entire history, semiconductor manufacturing has depended on optical lithography, which projects light through stencil-like masks to delineate, layer by layer, the infinitesimal structures that make up the transistors of an integrated circuit. But we’re fast approaching the point where optical lithography cannot take us where we need to go next. Consider that the transistors in next year’s state-of-the-art chips are so small that 4 million of them will fit into the period at the end of this sentence. The wavelengths of light we are using now are simply too large to print such fantastically dense patterns. We may not be at the end of the road for optical lithography, but it sure is getting tough to navigate.
For at least 25 years, lithographic researchers have anticipated the waning of optical lithography. They spent billions of dollars developing exotic lithographic systems that exploit radiation other than light: X-rays, electron beams, even ion beams. None became commercially viable, although there are always new contenders waiting in the wings. The most probable successor to optical lithography is extreme ultraviolet (EUV) lithography, which uses light of 13.5-nanometer wavelengths. But a little over a year ago, the experts realized that EUV would also fail to materialize before 2011 or 2012—the time frame during which chip makers will need a major lithographic advance to keep Moore’s Law going.
Fortunately, another option has emerged. It’s called double-patterning lithography, and it promises to extend optical lithography’s useful life for about four more years, or through two more doublings of chip transistor density. It will be a timely and lucrative reprieve: the Moore’s Law paradigm, which helped propel the integrated-circuit industry into the $255.6-billion-a-year juggernaut that it is today, will live to fight another day.
Double patterning is another of the many ”cheats” that lithographers have had to invent over the past decade to keep pushing the size of transistors into ever more remarkably minute realms. The technique involves complicated methods of doubling up the layers of printing, which means it’s about twice as expensive as conventional optical lithography, and it ties up the equipment for longer periods. But it’s the only method that will be able to tide the industry over until the arrival of EUV lithography in four or five years. Double-patterning lithography, to borrow a phrase from Winston Churchill, is the worst method that’s out there right now, except for all the others.
Doubling transistor density on a chip means shrinking its dimensions by about 30 percent. The industry is understandably desperate to see the pace of Moore’s Law continue, and that pace is dependent on the technology that can create those ever-shrinking transistors: optical lithography, also known as photolithography.
Photolithography literally prints microchips layer by layer. The technique’s most basic parameters are resolution and cost, and they are in more or less direct conflict. To print the billions of tiny individual features that make up a modern chip, you need extremely fine resolving power. And because that modern chip with nearly a billion transistors sells for only a few dollars, the printing method has to be stupendously cheap. Chip makers are constantly jockeying for advantage by trying to introduce new technologies ahead of their competitors, but for the most part they all move in lockstep between what are called technology nodes.
A ”node”loosely refers to the width of the smallest features of an integrated circuit—for example, the length of a transistor’s gate. In 1971, those 2300â¿¿transistor Intel 4004s were manufactured using technology that could create features measuring 10 000 nanometers (10 micrometers). Today’s most advanced chips are at a 45â¿¿nm node, ostensibly because the smallest features in the pattern measure 45 nm. Intel expects to begin producing 32-nm node chips in 2009. Chips based on 22-nm node processes are already under development and slated for production from 2011 through 2012. Using smaller wavelengths and larger lenses, the semiconductor industry has done a stunning job of scaling down transistors. Consider that if the transistors in the Intel 4004 had been the size of Humvees and had been scaled down to the extent that they have, they would today be as small as sesame seeds.
Every chip starts its life as a tiny patch on a gleaming round wafer of silicon about the size of a dinner plate. This wafer moves in and out of a series of machines through a fabrication plant the size of a football stadium. The result is a wafer imprinted with patterns of hundreds of identical microchips, which are then sliced and diced and go out into the world to populate routers, coffeemakers, ATMs, laptops, and fighter jets.
Optical lithography, which imprints the patterns onto the wafer, is a lot like old-style film and chemistry photography. It actually works a lot like a slide projector, in which a light source shines through a pattern to beam an image onto a surface.