Do you think Moore’s Law—the doubling of transistors on a chip every couple of years—will always hold? Think again. And find out what other common assumptions need to be revisited and revised:

In the semiconductor industry, several statements have gained such widespread acceptance that they have become conventional wisdom. But these famous assumptions, upon closer examination, don’t always hold up. In short, they’re more misconceptions than statements of fact. Below, we reveal the inaccuracies or inconsistencies behind some common beliefs:

Myth #1: Moore’s Law will continue to hold.

Sure, it has been accurate for nearly 30 years. But Moore’s Law—which maintains that the number of transistors per integrated circuit should double about every two years—may need to be revised. In fact, Intel cofounder Gordon Moore, who made the legendary observation in 1965 about the exponential growth of transistors on a given amount of silicon, recently admitted that he may have to change his statement to predict a doubling in circuit density only every four or five years. This is because chip advances are getting much harder to come by. And according to a new research paper from Intel, semiconductor makers won’t be able to reduce the size of transistors much, if at all, beyond 2021 because by that time, they would have reached the nanometer (billionth of a meter) scale and hit a fundamental constraint.

Myth #2: More integration is always a good idea.

Wrong. It’s not always better to pack millions of additional transistors and circuits onto increasingly dense chips. While that’s one road to more powerful chips, it could present problems. By using the latest technology to integrate more transistors, semiconductor makers could be overtaking customers’ needs, increasing costs and thus restricting the potential market for their products. Already, most companies find it challenging to fabricate chips well below the 100 million-transistor level. Cramming more than one billion transistors onto chips—while feasible—would dramatically drive up costs. Thus, increased chip integration is not an end goal but must always be weighed against technical and economic concerns.

Myth #3: CMOS is superior to more exotic processes.

Many believe that CMOS (complementary metal oxide semiconductor) technology—which is widely employed in chip production today—is always cheaper and better than less commonly used processes such as silicon germanium (SiGe) and gallium arsenide (GaAs). But while the technology has a firm grip on the chip industry (in 2003, SiGe and GaAs compound semiconductors accounted for only an estimated 1.5% of the world’s $140 billion chip sales) and is light years ahead of alternative processes in terms of investment and infrastructure, it’s losing ground to them in more and more applications. For example, when it comes to cell phone power amplifiers, chip makers are seeking the high speed of GaAs; the one-two punch of speed and power efficiency provided by SiGe; or the greater output current of bipolar CMOS (BiCMOS), which integrates CMOS with bipolar transistors. Additionally, the cost of leading-edge CMOS technology is skyrocketing. In short, CMOS is not always king.

Myth #4: Component scarcity is only a problem during upturns.

Many think that electronic components can only become hard to come by during upturns when demand for parts is soaring. But component scarcity can occur during downturns as well. Take the most recent one, for example, when many buyers at large OEMs (original equipment manufacturers) had a hard time finding parts because demand outstripped supply. In particular, purchasers struggled to find older semiconductors based on mature, but still viable, technology. This was because many suppliers either cut back on production of parts or stopped fabricating the components altogether. In short, parts are obsoleted all the time—not just when the industry is growing.

Myth #5: Procuring third-party intellectual property (IP) guarantees time and cost savings.

Contrary to popular belief, semiconductor makers can’t always count on saving time and money when they build chips from predesigned circuit blocks—such as memory, processors and standard interfaces—procured from outside specialists. While the strategy can indeed free up labor, money and time, letting chip makers focus on design areas promising the most added value, the approach can also bring unexpected headaches. For example, when trying to integrate third-party intellectual property (IP), some have encountered bugs or been misinformed about the capabilities of certain circuit blocks, causing them to expend extra money and engineering effort. One Silicon Valley chip executive, who did not want to be identified, tells Electronic Business, “many people don’t realize that almost no IP is plug-and-play. It requires some amount of work on the engineer’s part to modify it to work in your specific system. The lesson I learned is that you have to be skeptical.”

Sources:

Hit or Myth?
Russ Arensman
Electronic Business, February 1, 2004
www.reed-electronics.com/eb-mag/article/CA376457?industryid=2116

Intel Scientists Find Wall for Moore’s Law
Michael Kanellos
CNET News.com, December 1, 2003
news.com.com/2100-7337-5112061.html?tag=nefd_lede

Where Are the Parts?
James Carbone
Purchasing Magazine, December 11, 2003
www.purchasing.com