EDINBURGH, Scotland A steady rise to power of third-party developers of intellectual property (IP) will bring pressure on ASIC vendors, which have heretofore been the traditional leaders in system-level integration, according to two keynote speakers at the IP99 Europe conference and exhibition.
At the IP99 opening here Monday (Nov. 1), Wally Rhines, president and chief executive officer of Mentor Graphics Corp., alluded to a shift in power away from ASIC suppliers and other companies that do not try to proliferate intellectual property cores as widely as possible.
Then on Tuesday, Roger Fisher, senior director of marketing at Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC), addressed the "disintegration" of the semiconductor industry. Fisher said independent IP companies will inevitably rise alongside traditional semiconductor companies, fabless semiconductor companies and independent wafer foundries in an increasingly layered and co mplex electronics industry.
Rhines used historical data from the electronics industry and elsewhere to show that the relationships between various parameters and accumulated volumes, when plotted with a logarithmic scale on both axes, are remarkably linear, predictable and hold true across their industries.
"The IC learning curve is a 70 percent slope . . . if you double volume you reduce cost by 30 percent," Rhines said. He showed that similar relationships are true for quality learning curves portraying the ability to drive errors out of manufacturing or bugs out of software.
Rhines classified companies that supply IP cores for all design environments and make them available from all manufacturers as "winners." The losers are those that supply IP tied to a single manufacturer or a proprietary design environment, he said, describing but not naming conventional semiconductor companies and ASIC houses.
Rhines argued that the learning curves associated with quality, reliability and cost would favor IP suppliers that get their IP reused soon and often. "If you want to be unique and differentiated, sure, keep it in-house, but recognize that it will cost more to get down the learning curve," he said.
Rhines referenced ARM Ltd. as the classic example of a winner. Through partnership with multiple semiconductor companies ARM has put its microprocessor cores through multiple wafer fabs and achieved high market penetration. Other companies will need to follow this approach or risk falling behind on the various learning curves that apply to IP cores, he said.
"The cores with the largest number of reuses get the best quality, best reliability, best time-to-market and lowest cost," Rhines said.
Fisher was even more aggressive in his keynote on the following day.
"Our semiconductor world is disintegrating," he said. "It is inevitable so we must embrace it."
In an industry history described by Fisher, the semiconductor industry has undergone a continual process of layering from its earli est days, when a semiconductor company took in sand and shipped out packaged parts, to the current state of affairs, where three or four separate industries take responsibility for different parts of the IC production process. "Independent IP vendors will have a broad impact but especially on ASIC vendors," he said.
"Systems companies say, 'I love my ASICs,' but they want to go to customer owned tooling where they own the mask set and take responsibility for the design.
"Why? They want choice," he said: "Choice of IP, choice of manufacturing, protection of their own IP, control, and overall cost reduction."
Fisher went on to say that the advent of semiconductor manufacturing on 12-inch wafers in fabs that will cost $2.5 billion represents another milestone in the disintegration of the semiconductor industry and the "end-game for the integrated device manufacturers.
"One reason IDMs have continued to date is that they had leading-edge process technology. This year TSMC and one other foundry have caught up with the IDMs and the SIA road map," he said.
But that will lead to a significant shift. "Next year TSMC will be number two in wafer volume to Intel, so why would anyone go to an IDM and why would an IDM choose to stay in the game?"
Based on turnover, only Intel and a couple of the world's other leading semiconductor companies will be able to justify the building of a $2.5 billion 12-inch wafer fab, Fisher said. He did not say why foundries, with much smaller turnovers than IDMs, will be able to justify such fabs, but concluded, "The others are looking at partnering but I believe they will eventually turn to foundry."