Network Computingwww.networkcomputing.com

How could we create a list of the past decade's greatest technologies without recognizing the basic fuel that drives our industry? Advancements in IC (inte-grated circuit) technology throughout the '90s were nothing short of astonishing.

What we once would have thought of as a supercomputer now sits on our desktops. Switches and routers running at blazing speeds would have been impossible at the outset of the decade; today, not only are they a reality, but they often cost less than $100 per port. Take a look at a 1990 high-performance network interface card, and you'll find a dozen or so ICs and a hefty price tag. Today, you'll see NICs that perform better, cost far less and use only a couple of chips (rendering them far more reliable to boot).

Even the most technical among us can't help but marvel at how chips get faster, more complex--and smaller--year after year after year. It's no easy feat. Through a combination of process improvements, design-tool advances and research breakthroughs, the IC industry does its best to continue to live up to Moore's Law.

In a now-famous observation made in 1965, Intel co-founder Gordon Moore noted that the density of transistors on ICs had been doubling every year or so, and he predicted that the doubling would continue for the foreseeable future. While the doubling period has slowed to about every 18 months, IC manufacturers still use Moore's Law as the benchmark to gauge their progress. But wait a minute--Moore talked about transistor densities (often referred to as "gate densities," with gates being patterns of transistors); what about speed? As it turns out, every time transistor size shrinks, the transistors can turn on and off that much faster. Thus, CPU clock rates have doubled almost every two years. Finally, as gate densities increase, designers try to reduce the amount of power chips consume, and consuming less power is always a good thing (as laptop, PDA and cell-phone users will agree).

The goals of chip technology are straightforward, but nature's impediments to those goals are monumental. As transistors get smaller, some properties of commonly used materials within chips start hurting their performance--sometimes severely. Breakthroughs in ways of using new materials with better electrical properties help chip makers push the limits. For example, while aluminum has been widely used for gate interconnections, IBM found a way to use copper, which is a better conductor and therefore allows interconnection lines to be smaller and run faster. Similarly, advances such as silicon germanium and silicon-on-insulator transistors (also from IBM) help push gate sizes down.

Networking has benefited the most from advances in ASICs (application-specific integrated circuits), also known as ASSPs (application-specific standard parts). These devices are at the heart of switches and routers, and are responsible for the high speeds and low price points we now enjoy. ASICs improve performance by implementing complex protocols directly in hardware, processing data many times faster than could general-purpose chips via software algorithms.

As chips have evolved from housing hundreds of thousands of gates to many millions, creating and verifying chip designs have become commensurately more complex. Software tools have progressed in tandem with chip technology to help designers harness the incredible power of today's ICs. How long can this amazing balance of process, research and tool improvements continue to satisfy Moore's Law? The common wisdom among those who should know (including Moore) says we have at least another two decades of steady advances ahead of us.

-- Mike Lee and Art Wittmann