That has long been the mantra of semiconductor makers and researchers, who while still mired in the tech funk, manage to come up with advances that promise to "revolutionize" the technology. However, meaningful semiconductor advances have been, and will continue to be, on an incremental track.
IT buyers and bioinformaticians hoping for a chip that will help speed the processing of biological, chemical, and clinical research data should bear in mind that sea-changing leaps are very rare.
"There will be things on the bleeding edge, the leading edge, here tomorrow and here today. Everything passes through four stages. The roadmap of the future has continuity that moves you forward year after year," says Bernie Meyerson, chief technology officer for IBM Corp.'s technology group. "But let me be honest. Very few proposals are truly game-changers. The thing you have to look for is when something moves from the bleeding edge to the leading edge."
Consider that chipmakers cooperate with each other via a multitude of standards bodies, including the International Technology Roadmap for Semiconductors. "Semiconductors are a $150-billion industry. You're not going to get things that just come out of the blue," says Fred Zieber, president of Pathfinder Research Inc.
An example of a bleeding edge technology is superconductivity — the lack of resistance to electricity at very low temperatures. Discovered in 1911, it could revolutionize chipmaking were it not for the prohibitive economics of cooling a superconducting circuit. Other "revolutionary" technologies that fizzled or were relegated to small niches include IBM's Josephson Junction and chips made from gallium arsenide.
"Superconductivity is the bleeding, bleeding edge," says Meyerson. "Can you run with it forward in time and through the various life cycles? That's the motion you have to watch. Look for movement off the bleeding edge." When bleeding-edge advances in semiconductor technology first hit scientific and technical journals, read them for entertainment purposes only.
Nanotechnology is all the rage, but a nanometer (one billionth of a meter) in this context is just a gussied-up measurement to gauge the size of a transistor, which until recently was measured in microns (millionths of a meter). Now that transistors are below 0.1 micron, they are measured in nanometers.
Nanotechnology sits squarely in the leading-edge and here-tomorrow categories, but with an eye on here-today. On June 10, IBM announced that it is teaming up with Cisco Systems Inc. to produce Application Specific Integrated Circuits (ASICs) with 90-nanometer transistors (0.09 micron, as compared with the last downsizing notch at 0.13 micron). Two months earlier, Taiwan Semiconductor Manufacturing Company said it would start producing samples this fall of 90-nanometer transistors.
Intel Corp. demonstrated 90-nanometer technology in March, according to Intel technology analyst Rob Willoner.
"The first obvious benefit of .09 [micron] technology is better transistor density. You can get the same number of transistors in half the area [of .13 micron]," says Willoner. "When there's more functional blocks on the chip, you can add graphics, security, and bigger caches. The engineer can play with the added transistor budget. This all gets you closer to system-on-a-chip." (Intel does not have an system-on-a-chip on the market yet.)
For instance, the Pentium 4, built with 0.13-micron technology, has 55 million transistors. The static RAM shown by Intel in March has 330 million.
How the size of a transistor is measured is something only an electrical engineer can appreciate. It's not as simple as gate length or the corner-to-corner measurement of a TV screen. "It's the minimum distance between the center of parallel interconnects divided by two. Even that definition is not precisely adhered to," Willoner says.
The latest size reduction is significant only because transistors have dipped below 0.1 micron. Of course, they've been steadily getting smaller since they were invented in 1946 and started to appear on circuit boards in the '50s. (What 50-something can forget a "six-transistor portable radio?")
Less is Moore
Semiconductor evolution started in earnest in 1965 when Intel co-founder Gordon Moore predicted in a seminal paper appropriately titled "Cramming more components onto integrated circuits" that chip capacity would double every 18 months. Intel's microprocessor history began with the introduction of 4004 in 1971, which was downright obese at 10 microns. That's more than 100 times bigger than 90-nanometer technology.
"I think back two, five, and 10 years and everyone asked when we would get to one micron — and now we're at .09," says Markus Levy, senior analyst with Cahners Micro Design Resources. "We're all headed in a direction where chips will be so dense, they can emulate the human brain. When do the laws of physics come [into play] and it's not possible [to get smaller] anymore? A transistor could be a couple of atoms."
Already, Taiwan Semiconductor has built a computer model of a 9-nanometer transistor and Intel says it has actually built a 15-nanometer transistor. Researchers at Princeton and Stanford universities have developed a new 10-nanometer manufacturing technique. But problems with electrical current leakage, heat, design, chip yields, and packaging rise exponentially at these sizes.
"It's a terrible disservice to go out and brag about the first 90-nanometer technology," cautions Meyerson. "You can always make it smaller. The problem is the impact on fundamental characteristics of the device. There is where the excitement begins. They become so small that the behavioral physics of that region changes. When you shrink to the next generation, problems go up exponentially. If you make [insulation] half as thick, its conductivity might not just go up by two. It might go up by 1,000."
To make the point, Meyerson compares a glass of water, which looks like what we think of as water, to a molecule of water. "That molecule doesn't look anything like water, but it is water."
The remarkable track record of Moore's Law could slow down as these behaviors intensify, according to Taiwan Semiconductor's Director of Brand Management Chuck Byers, who says 10-nanometer technology is probably 25 years out. "It'll [arrive] well after you and I retire."
Maybe so, but products are already being built as small as one nanometer. The remarkable electrical, mechanical, and chemical properties of nanotubes — tiny honeycomb tubes fashioned from carbon — make them useful in futuristic applications as diverse as artificial muscles and miniature electronics.
Many of these applications, such as nanomachines that could be a single nanotube robot or actuator, are probably decades away. Still, there is little doubt this technology will find its way into semiconductors in the biosciences eventually.