ARM, founded in 1990, is the best-known example of an intellectual-property (IP) cores company, one that makes its living not by manufacturing and selling chips but by licensing them. Other semiconductor companies pay fees to use ARM's designs and architectures, and then pay royalties on any chips sold that are built around those designs.

Its story has something of the aura of a founding myth in the gold rush that the IP industry has become. Launched as a joint venture among Acorn Computers, VLSI Technology and Apple Computer, ARM has cut the trail for a host of other IP purveyors, such as DSP Group Inc., Rambus Inc., ARC Cores Ltd. and many others.

With hindsight, it seems clear that this revolution was a logical step in the maturing of the semiconductor industry. But in 1990, success for Advanced RISC Machines Ltd., as it was then called, was not assured. Nor was the business model created out of any sense of historical imperative.

As Robin Saxby, chairman, president and chief executive officer of ARM, tells it, Advanced RISC Machines became a "chipless" chip maker by default. Indeed, the intellectual property ARM inherited from one of its founding parents, the 32-bit Acorn RISC Machine, had languished in VLSI Technology's data book as a functional block for ASIC design for several years in the 1980s without gaining much interest in the industry.

When he was brought in as the head of ARM Saxby decided that wisdom lay in avoiding the costs incurred in sales and marketing. He reasoned that partnering with semiconductor companies was a better option, since they would bear those costs and deliver chips based on ARM cores, while his staff of 12 engineers got to concentrate on what they were good at: engineering.

From that starting point Saxby has made a tremendous contribution to the development of the semiconductor industry-not to mention his shareholders' wealth-by using the transportability of gate-level, register-level and architecture-level machine-readable descriptions to turn chip design into a software business. He has effectively created a chip business with a cost structure somewhat lower than that of a traditional software company.

ARM was not the first to exploit the value of its design work in the form of intellectual property, however. In the 1980s Texas Instruments Inc. had become aggressive in negotiating the licensing of its strong fundamental patent position, and was building a considerable revenue stream from this IP.

"About the time [the late '80s] Texas Instruments made more in one particular year licensing its patent portfolio than from semiconductor sales," said Mark Jones, who in 1976 founded a small design services company, MEJ Ltd. (Guildford, England), that also pioneered the fabless and chipless business model. "We [at MEJ] didn't necessarily see an IP connection, but we realized that licensing designs was more profitable and more scalable than design services."

Like ARM's Saxby, Jones-who is now vice president and general manager of IP business at memory architecture company Mosys Inc. (Sunnyvale, Calif.)-holds that the move into IP licensing was more opportunistic than premeditated.

"Back in the 1980s, as a design services company in the U.K., we were doing a lot of work for one company, Amstrad," he said. "At one point the work turned down and I had a lot of engineers and not a lot of work for them to do. So we started work on megacells that we expected would be useful later on-it was a way of getting ahead and building a library, really for our own design reuse.

"We started with things like interrupt controllers and moved on to an 8051-style microcontroller and developed quite a lot of these small cores," Jones recalled.

The philosophy of design reuse at this higher level of abstraction was already known at this time, but was offered almost exclusively by the gate array and standard-cell ASIC vendors as an addition to their libraries of basic gates.

"Our design-service customers started getting interested in our macros but it was NEC that was the first to ask us if they could license them in about 1988 or 1989," said Jones.

To follow its IP licensing strategy, MEJ was renamed 3Soft Ltd. and Jones moved the company to San Jose, Calif., as 3Soft Inc. It was acquired by Mentor Graphics Corp. in the mid-1990s.

Looking back from his position with memory-IP licensor Mosys, Jones said, "I think the rise of the IP company was absolutely inevitable. We can see that the history of the semiconductor industry has been one of vertical disintegration for 20 years. There's an evolution process going on here that is literally the survival of the fittest."

In Jones' view, "The survivors will be those that focus, and they will tend to do it better than broader companies. An IP company will tend to be better at creating, supplying and supporting a circuit design than those companies that are distracted by having to fill a fab."

Of course, a circuit design is of no practical use until it is implemented in silicon, admitted Jones. "Hence the rise of the foundry semiconductor companies. They focus on manufacturing and tend to do [that] better."

A few integrated device manufacturers, such as Intel, IBM Microelectronics and Lucent, still lead the way in process technology development, but the independent foundries are close behind. And in business terms, the foundries have a broad customer base that provides greater flexibility in filling fabs, which today can cost $2 billion to build.

"I think the true significance of the IP revolution was missed in the early days," said Jones. "There's still some people saying 'it will happen,' but most of us now realize it has happened and it's a question of execution."

With better backing and a key piece of IP in house, the original Acorn RISC Machine, ARM was always likely to make a bigger impact than MEJ. And so it has proved.

ARM's original plan was to let VLSI Technology and one other chip company, GEC-Plessey Semiconductors, manufacture ARM processors for Apple's Newton, an early handheld personal digital assistant. Saxby is credited with seeing that the semiconductor partnership with Apple would need to be extended further to build ARM's reach to a level that made sense in business terms.

The fact that the Newton was failure in the market was, if anything, a boon to ARM, giving the company the incentive to attack and eventually dominate the market for embedded controllers in mobile telephones.

Saxby says the automotive market is the company's next target. "We expect automotive applications to really take off next year [2000], although the first application might be a motorbike," he said. But Saxby's ambitions range far beyond automotive and telecommunications sectors. "We want to be in everything digital by 2010," he said flatly.

By then, semiconductor manufacturing should be able to deliver multiple systems on each individual chip, and computer architecture will have probably gone through several changes to embrace parallel processing and other techniques. And ARM will naturally migrate to embrace architecture and software IP.

The success of ARM notwithstanding, Saxby believes it will not be easy for newcomers to break into the IP business. "I believe a limited number of IP companies will be successful. There are a few key areas-memory architecture, CPU architecture and DSP architecture-that will sustain IP companies," he said.

ARM's latest deal, with Intel Corp., is a five-year architecture licensing agreement that supports Intel's design and manufacturing of StrongARM processors and carries IP licensing to new heights of sophistication. In effect ARM has licensed Intel with a series of architectures that have yet to be defined.

The payment terms remain confidential, but as one senior executive at ARM pointed out, "It's a five-year deal licensing architectures to Intel as they become available. But we are in the business of selling licenses."

On the strength of partnerships like that, perhaps it makes sense that that the highly focused ARM is a more valuable business than the asset-laden British Airways. Indeed, in the next millennium perhaps ARM, with relatively few but very bright engineers and its focus on IP, is in a better position than even Intel, which has thousands of engineers and the heavy costs of process technology development and wafer fabs to build and keep fed with designs.

The Century of the Engineer: Misunderstood Milestones