The 5 most enduring principles
In the beginning, there was Moore's Law, and from it come the industry's fundamental principles.
By Bill Roberts, illustration by Chuck Mackey -- EDN, November 1, 2005
|
Brian Halla, CEO of National Semiconductor, first heard it as a young engineer around the Intel watercooler in 1975, the year Electronic Business launched. Aart De Geus, CEO of Synopsys, first heard it at a technical conference in 1979, when he was still a graduate student. T. J. Rodgers, CEO of Cypress Semiconductor, also first heard it in 1979, from the horse's mouth—Gordon Moore himself.
What each heard was the predicted demise of Moore's Law. The very idea bemuses them now, because it has been the industry mantra and self-fulfilling prophecy for four decades. In an article in the April 19, 1965, issue of Electronics, written three years before he and Robert Noyce left Fairchild Semiconductor to found Intel, Moore predicted that the number of transistors on a silicon chip would double each year. A decade later, he recalibrated his law to every two years, where it remains.
Although the industry has constantly changed over 30 years, Moore's Law and a few other fundamental principles have been as consistent and solid as bedrock. Like tectonic plates, they move but do not shift radically. Earthquakes tend to be caused by so-called paradigm shifts: the shift in focus from space, military and mainframes to desktops; from the desktop to networking within the walls of the enterprise, to the World Wide Web, to connecting everything outside the walls; and now the proliferation of consumer products, especially the multifaceted cell phone.
With the help of executives, technologists and other observers, in interviews and e-mails, EB editors sized up these 30 years and concluded that industry bedrocks include increasing specialization, strategic globalization, continuous innovation and making money.
They all rest on the sturdy foundation of Moore's Law.
The law was already 10 years old when Cahners Publishing converted Electronic Purchasing magazine to Electronic Business, in October 1975 (Reed Publishing acquired Cahners in 1977). Like the executives interviewed, generations of EB editors have heard dire predictions of the law's demise.
Someday, someone is likely to be a prophet.
Moore's Law
Change is constant, a fact Moore's Law codified and calibrated for electronics. It became the industry's equivalent of Satchel Paige's advice: "Never look back. Someone may be gaining on you." It saved electronics from the complacency that has caused many industries to grow fat and lazy. Companies that broke the law are no longer with us. "If you tried to apply the same kind of law to any other industry, you'd have a Rolls Royce anyone could buy for a quarter, drive downtown and throw away," Halla quips.
Moore's Law transcended what it was at first, a statement of physical possibilities, observes Satoru Ito, CEO of Renesas Technology. "Because of Moore's Law, the industry has had a common road map for technological innovation. This allows partnerships and planning for investment."
Aki Fujimura, CTO of Cadence Design Systems, agrees: "The planning ability it gives the entire industry is unlike anything else. It is the pulse against which the entire ecostructure of the semiconductor industry can pace itself and coordinate itself."
Ito also points to the law's cyclical self-fulfilling nature: "Advances in technology create new applications that cause the market to expand and demand to grow. This growth, in turn, fuels new advances in technology."
Moore's Law became an economic barometer, closely watched by Fed Chairman Alan Greenspan, who often alludes to it in his statements on the health of the economy. Geometric scaling turned into economic scaling, driving more power, faster and cheaper, and leading to more integration of more functions in a smaller space. No economic sector was untouched by the productivity improvements.
"If you tried to apply the same kind of law to any other industry, you'd have a Rolls Royce anyone could buy for a quarter, drive downtown and throw away."
—Brian Halla, National Semiconductor
"Where else do you find an industry that can reduce its prices for a given level of function or performance each year?" says Dennis Monticelli, chief technologist at National. "This is possible because we can build more transistors in each batch lot than in the previous two years. This translates into ever higher levels of chip integration."
That last point is worth reemphasizing. "Integration trumps everything," declares Andy Rappaport, a venture capitalist at August Capital, which invests in dozens of chip startups. "Achieving low cost, small size and low power is almost always a function of integration."
Almost as soon as Moore laid down the law, skeptics began to prognosticate its end.
Rodgers heard Moore himself cast a pall over it during a 1979 board meeting of American Microsystems Inc., where Rodgers was a product manager. AMI was negotiating with Intel for some technology. During full disclosure, Moore described a physical limit Intel had encountered. "He postulated that it might not be able to solve the problem," Rodgers recalls.
Around that time, Intel had just introduced its 8088 processor, with 29,000 transistors, which was adopted a couple of years later for the first IBM PC. In contrast, Intel's most recent processor, the Itanium 2, launched in 2004, has 592 million transistors (see the chart "Moore's Law at Work," below).
In 2005 one certain threat to Moore's Law is the difficulty of designing chips with that many transistors without breaking the bank. If history is any guide, the industry will once again be forced to rely on its ability to specialize—which is, not coincidentally, another one of its bedrocks.
Specialization and Partnering
With the number of transistors on a same-size die doubling every two years, the day inevitably came when a single company could not marshal all the resources needed to do everything: develop all IP, build design tools, design integrated circuits, manufacture chips, distribute, design systems and assemble components on boards or boards into products. Moore's Law virtually mandated increased specialization. This, in turn, led to new business models, including those relating to chip foundries and contract manufacturing. The vertical company went the way of the vacuum tube (see the chart, "The Long March of Specialization," below), and companies that did not learn to partner joined those that broke Moore's Law.
Specialization works only if partnering is successful. "Partnering is the air we need to breathe," says Andrea Cuomo, executive vice president of STMicroelectronics. "A large network of strategic alliances built from customers, suppliers, research partners and even competitors contributes more than revenues. Alliances also help a company access and assimilate complex knowledge from disparate sources."
Moore's Law begot complexities that demand partnering earlier in the flow. "We have been continually moving upstream," says Rick Wallace, president of KLA-Tencor, which develops yield management technology for chip manufacturing. KLA first collaborated only with foundries; about 10 years ago, it was pulled into product development. In the past two years, it has also been increasingly involved in design.
As the matrix of partners has expanded—including even competitors—partnering has become more challenging. Cuomo believes that companies have to work hard at it and that success "is a function of corporate culture and past cooperation—even of personal relationships and entrepreneurial culture diffused throughout the organization."
In 2005 design productivity demands better partnering. Designs that used to require 20 engineers now need 100. The industry is not solving the problem fast enough. "Profits are slipping, because of the inability to scale productivity to fully leverage the advanced technology," says Rajeev Madhavan, CEO of Magma Design Automation. He estimates that it costs $80 million or more to design an SoC. "You can never recoup that investment in most cases."
"Because of Moore's Law, the industry has had a common road map for technological innovation. This allows partnerships and planning for investment."
—Satoru Ito, Renesas Technology
Mark Pinto, CTO of Applied Materials, agrees: "Design paradigms need to change. Nothing else will be revolutionary. Alliances will have to get stronger to deal with [design productivity problems]. We're certainly working with EDA to try to get there."
Some executives complain that the EDA companies are too busy suing each other to work together to solve the design productivity problems Moore's Law creates. It is a bit unfair to fully blame EDA companies, given the litigious internecine battles in other industry segments over the years. Still, some EDA executives agree.
"We're way behind," Madhavan observes. He faults EDA for not working more closely with the rest of the industry until recently. "Our relations with customers are much closer today." Madhavan, whose company is engaged in a patent battle with Synopsys, also says there's validity to the complaint that EDA companies have not learned to work together: "Part of the problem in EDA is the not-invented-here syndrome. The total market hasn't grown very fast, so there has been room for only a few big fish in this pond. As a result, partnering between the big EDA companies and the rest of the industry has been lacking. The closest thing we have to coopetition is acquisitions. The EDA industry's tactics serve only to stall delivery of the productivity increase that customers so desperately need."
Synopsys' De Geus counters, "We are working with parts of the industry we've never worked with before. We interact much more now with the fab side. And we track much more information between equipment than before." He recognizes that customers need interoperability among competing products and adds that Synopsys has worked on creating industry standards for years.
Globalization
Moore's Law has spawned complexity, specialization and partnerships. These trends, in turn, spawned globalization, driving companies to look abroad, at first for cheaper labor and later for a bigger pool of scientists, engineers and partners. Operating around the world also created the ability to work on designs around the clock and to develop new markets.
"Globalization has always been a big part of electronics," observes Gary Smith, a Gartner analyst. "We've been expanding globally almost from the beginning."
Global resources became a necessity, ST's Cuomo argues. "To succeed, a company needs to identify new sources of relevant technology; bring this new knowledge into products and market opportunities; and translate all of this into efficient, flexible and cost-effective manufacturing. In this view, globalization is strategic. In addition to providing lower-cost labor, emerging markets are sources of fresh insights and innovation."
"A partially filled fab loses a tremendous amount of money, so when you build a new fab, you want to fill it as fast as possible. Moore's Law compounds the problem, because you have so many more transistors to design at each process node."
—Gary Smith, Gartner
Globalization no longer just means having sales offices and some plants abroad, Michael Marks, CEO of Flextronics, says. "To be successful today, you have to tap into talent everywhere. China is best for manufacturing. India is best for software. Ireland is best for call centers. We have skill sets in different places. We have people all over the world. If you don't take advantage of the best the world has to offer, you're a loser. That's really new."
If globalization is viewed strategically, it does not always make sense to move offshore. "China is still too costly for manufacturing complex, low-volume products such as semiconductor manufacturing equipment," says Pamela Gordon, president of supply chain consulting firm Technology Forecasters. "Executives have proven again and again to be shortsighted, seeking out the lowest wages and not taking into account all other factors." Her point: Globalization must be done for strategy reasons, not just for cheap labor.
Innovation Forever
Electronics entrepreneurs already existed before Moore's Law, but they certainly benefited from it. In Santa Clara County, which became known as Silicon Valley, Hewlett-Packard was founded in 1938 and World War II spawned many defense electronics companies. William Shockley opened Shockley Semiconductor Laboratory, the first chip company, in Palo Alto in 1955. He had already invented the solid-state transistor at Bell Laboratories in Murray Hill, N.J., and chose Palo Alto for many reasons: He was financed by Beckman Instruments, in southern California; he could recruit engineers from Stanford; and his aging mother lived in Palo Alto.
Put off by Shockley's autocratic style, eight engineers, led by Noyce, left to start Fairchild Semiconductor in 1957, with a new kind of investor. Teaming up with Arthur Rock, a young stock investor, Noyce got money from the first venture capital fund. Noyce and Moore later founded Intel, with funding lined up by Rock. Noyce proved twice that a smart engineer with a better idea could find venture funding and start a company.
The rest is history. At least 25 chip companies trace their lineage directly to Shockley, and dozens more indirectly. Thanks to creativity, a risk-taking mentality and the predictability of Moore's Law, Silicon Valley has launched thousands of startups, with immeasurable impact on the world's economy.
Moore's Law showed investors where the industry was headed and gave startups a predictable wave they could ride if they could figure out when, where and how to catch it.
"The potential of all this technology could never have been realized without the ingredients of a thriving entrepreneurial culture," says National's Monticelli. "Putting investment money at risk and putting careers at risk are hallmarks of this culture. Engineers and businesspeople from around the world have been attracted to areas that featured this culture, codified in an infrastructure of people, services, money and a good work ethic."
Startups benefited from and contributed to specialization. When building a fab became too costly, the foundry model took off, creating a new generation of fabless startups. When the cost of developing all IP became too time-consuming for established companies, startups sprang up to fill the need.
Profits, Profits, Profits
Investors and entrepreneurs were willing to risk money and career, because the potential rewards were vast. There is still money to be made in electronics, but it is never easy. "You have to feed the elephant," notes Gartner's Smith. "A partially filled fab loses a tremendous amount of money, so when you build a new fab, you want to fill it as fast as possible. Moore's Law compounds the problem, because you have so many more transistors to design at each process node."
No matter how geeky electronics seems to be, it is still driven by pure economics, argues KLA-Tencor's Wallace. "Innovations that find their way into mainstream production come about only because there is an economic need for them. Moore's Law isn't just about doubling device density; it's also about lowering the cost per bit. This is a capital-intensive industry that is becoming only more so, and ideas that don't demonstrate a clear ROI are abandoned in midstream or are destined to remain just ideas."
Wallace remembers when cost didn't matter as much. Early in his career, he sold chips to the defense industry, where quality and reliability were all that counted. Today, he says, "Our customer base is becoming more sensitive to more-general economic challenges and opportunities. Our customers have to be sensitive to the economics, and we have to step up the operational excellence."
"Moore's Law will die for lack of money, not because of some quantum mechanical problem. It will end because people will choose to stop investing. That will probably happen in the next five to 10 years."
—T.J. Rodgers, Cypress Semiconductor
Success equals good profit margins, says Scott Moody, CEO of AuthenTec, a startup that makes fingerprint-sensing chips. "There is nothing as bedrock as gross margins. I took a look at it a few years ago, comparing gross margin to company value. They track exactly." He looked at 20 chip companies and found that those with 50 percent gross margins or more had high stock market values; under 50 percent saw value drop precipitously.
August Capital's Rappaport says economic value is always a moving target; the spoils typically go to the first movers, not to those who improve things. "In my 30 years in the industry, this is the principle that has been the most consistently misunderstood or ignored."
Time on the Clock
If Moore's Law is the most important bedrock, how much longer can it underpin the industry?
Halla, who becomes president of the Semiconductor Industry Association in November 2005, thinks there are much larger problems than the future of Moore's Law. The entire innovative environment is threatened, he argues. "All these bedrocks you talk about have been pretty much shaken to dust. Government policies used to nurture innovation, and now we don't even have basic research" (see "The End of Innovation?" October 2004). He runs through the litany of issues that concern the U.S. industry, including research, fewer visas for grad students and engineers from abroad and onerous provisions in Sarbanes-Oxley.
But strictly speaking, what about Moore's Law? "As long as applications continue to need more transistors—and there is every indication that they will—investing in the road map to go to next-generation technologies such as 65 nanometer, 45 nanometer and beyond is the most economical solution," Applied's Pinto argues.
De Geus thinks Moore's Law is in trouble but that that just means more opportunity. He cites two sets of physical problems: power—both dissipation and leakage—and manufacturing yield. In 2002 he heard people predict that the transition to 90 nm was impossible. Today, many of his customers have moved to 90 nm and a few to 65 nm. He is not cavalier, but he says, "Technologists love tough problems, because they are the ultimate challenge."
Cypress' Rodgers agrees that the physical problems can be solved but believes that the law may become irrelevant. "Semiconductor densities will grow about another order of magnitude, and then at that point, if you want more memory, you'll just get more chips. It will be cheaper for systems designers to buy more chips than to continue to shrink."
Rodgers, who has heard the dire predictions for decades, now offers his own: "Moore's Law will die for lack of money, not because of some quantum mechanical problem. It will end because people will stop investing. That will probably happen in the next five to 10 years."
Someday, someone is likely to be a prophet.
Feedback question: What's your theory on the longevity of Moore's Law? Send your thoughts to feedback@eb.reedbusiness.com.
Bill Roberts has been working on a book this year. He will resume his regular contributions to the magazine in 2006.
MOORE'S LAW AT WORKhow the number of transistors on one microprocessor has grown at Intel
| SOURCE: INTEL | |
| 1971 |
2,300 |
| 1972 |
2,500 |
| 1974 |
4,500 |
| 1978 |
29,000 |
| 1982 |
134,000 |
| 1985 |
275,000 |
| 1989 |
1,200,000 |
| 1993 |
3,100,000 |
| 1997 |
7,500,000 |
| 1999 |
9,500,000 |
| 2000 |
42,000,000 |
| 2003 |
220,000,000 |
| 2004 |
592,000,000 |
