'

Does Moore's Law no longer apply?

Doubling processing speed every 18 months isn't the only way to build a better chip, IBM says

If Moore's Law doesn't hold up, will it matter? IBM says no.

The longstanding high-tech principle that processor performance doubles every 18 months came under question last September, when Intel scientist Paul Packan published a paper that stated chip engineers hadn't found ways around microprocessor design barriers for chips sets to be manufactured after this year. Since then, the Semiconductor Industry Association has predicted that chip performance will continue to increase over the next 15 years, but doubts about the physical limitations of silicon chips remain.

Moore's Law is named after Intel co-founder Gordon Moore, who in 1965 said a silicon chip's number of transistors, and therefore its processing power, would double every 18 months.

However, if Moore's Law has reached its limit -- and IBM doesn't necessarily think it has -- Big Blue says other things besides clock speed will work to increase performance.

IBM says it is taking a more holistic approach to its semiconductor research and development. Instead of focusing on clock speed, the company says it is deploying and developing a number of new processor technologies that do the job better.

IBM is utilising copper interconnects, for example, on its PowerPC processors. The company is also beginning to ramp up new silicon on insulator technology, and late last year it marked the first anniversary of the shipping of Gallium Arsenide chips.

While these technologies don't necessarily make chips faster, IBM says they do improve performance. For example, copper interconnects are more power-efficient conductors than the aluminum interconnects currently in wide use, meaning a chip will take less power and run at a lower temperature for the same clock speed.

And there are other considerations, such as input/output. A processor can be extremely fast, says Russell Lange, director of Silicon Technology Strategy for IBM's Microelectronics Division, but if it does not have data to process what good is it?

"Faster is generally better, but there are other values that are may be more valuable in terms of making information technology more useful," Lange said. "Do you need your PC to run twice as fast? Can you type twice as fast as a result of that?"

When it comes to the choice between having a notebook PC or a desktop PC, for example, Lange said, "I'd rather have the portability than three times the frequency. That's because I can take it with me."

Lange contends there are many dimensions to making information technology useful to people. Another way to increase performance is to improve memory capacity and improve its ability to move more data faster to a processor.

IBM this week announced new memory technology that increases the amount of data that can be stored in a memory chip. Lange said IBM has developed a way to break a memory density barrier -- or SF82 (squared) barrier -- by finding a way to completely immerse the memory trench, that part of its memory chip that stores data bits, in silicon and shrinking the size of the trench, Lange said.

Increasing memory capacity by making memory cell components smaller and increasing the amount that can be stored in a single chip makes it possible for IBM to build processors with larger amounts of integrated memory. Integrated memory cuts processing time because the chip can fetch data from close at hand, rather than having to reach out to the main memory of a system across a much slower bus. Bus speeds are lagging processor speeds by factors of four or five, a gap that isn't likely to close any time soon.

IBM is also working on improving network efficiency by implementing network processors, which it says can analyze data more quickly, for devices such as routers. These check the contents and destinations of packets through specially designed circuits rather than software, improving effectiveness, speed and reliability.

In terms of advances that do directly affect pure clock speed, meanwhile, Lange said Big Blue has developed chips with features as close together as 0.05 microns in its labs.

The current industry standard is 0.18 microns and will move to 0.13 microns next year and 0.10 microns in about 2003.

What do you think? Tell the Mailroom. And read what others have said.