San Jose, Calif. -- The further unraveling of the Crolles2 Al- liance, coupled with Texas Instruments Inc.'s decision to use foundry partners for its future process technology, had observers wondering last week whether the entire semiconductor industry was about to go fabless.
With logic processes fast becoming commodities, the real differentiator is in the IC design itself, not the manufacturing. So industry watchers speculated about which integrated device manufacturers (IDMs) might be next to shed their costly fabs altogether, or at least move toward a "fab lite" model, in which less manufacturing is done internally and more is sent to foundries.
"It is certainly the case that the IDM business model is out of fashion," said Didier Scemama, semiconductor analyst with ABN Amro Bank in London. IDMs closer and closer to the top of the chip market rankings are gradually relinquishing control of manufacturing, Scemama said.
"The complexity of 32-nanometer [manufacturing], with immersion lithography and double patterning, made it too dif- ficult for Texas Instruments," he said, adding that the manufacturing advantage apparently wasn't worth the development cost.
Indeed, "process technology can be bought instead of being developed today," said Doug Freedman, an analyst with American Technology Research Corp. (Greenwich, Conn.). "Clearly, the foundries are capable of providing leading-edge capabilities."
In what seemed to be another nail in the coffin of the troubled Crolles2 Alliance, members Freescale Semiconductor Inc. and STMicroelectronics Inc. separately announced plans to move in different directions and form alliances with other fabs or foundries. Freescale has joined IBM Corp.'s "fab club" (see story, this page), while ST plans to shift gears and align with "industry leaders" at the 32-nm node (www.eetimes.com, search article ID: 197000457).
Given the earlier defection of NXP Semiconductors, "Crolles2 is dead," concluded one analyst.
At the same time, TI stunned the market by disclosing it will drop the costly business of digital logic process development (see story, this page). TI said it plans to close its 200-mm Kilby wafer fab in Dallas and move the equipment into its existing analog fabs. The company will cut 500 jobs by year's end.
Other major IDMs are expected to revamp their own chip-manufacturing strategies. Analysts last week pointed to Cypress, IDT and Spansion as candidates.
With the cost of 300-mm wafer fabrication facilities approaching $5 billion or more, coupled with the soaring price of process development and IC equipment, the decision to work more closely with foundries appears to be a tough but necessary choice if IDMs are to rein in their costs while keeping up with Moore's Law.
Increasingly, IDMs are allowing the foundries to handle not only more of the traditional IC production requirements, but also the R&D. Over time, process technology will become less of a differentiator for many chip companies, at least on the digital CMOS side. IDMs will have little or no process technology advantage over their fabless rivals, many of which use the same foundries to gain access to cutting-edge technologies.
"The IDMs go from being the source of the process and in control of the relationship to a situation where they are just another customer," said Malcolm Penn, chief executive of market analysis firm Future Horizons (Sevenoaks, England). "At that point, the dynamics will change dramatically. When there are only a few sources of process and manufacturing, the price will go up. And if everybody has the same process, how do you differentiate--only in design, or only in the software running on the design? The differentiation all becomes second or third order."
It is possible that TI and ST still intend to build fab shells on an occasional basis to house licensed-in processes, so that they can continue to manufacture digital logic at the leading edge. If that is the case, the companies could be seen as moving to a "fab lighter" position.
But Penn insisted that process development and manufacturing are inextricably linked. The logical extension of the TI and ST decisions is that manufacturing of digital CMOS will gradually wither in the fabs, Penn said. And in his view, giving up process development is done at the long-term peril of shareholders.
"When chip manufacturing is controlled by a quasi-monopoly of three or four foundries. how are they going to play the game?" Penn asked.
Both companies intend to carry on manufacturing more-exotic processes, such as analog and RF on CMOS. But Taiwan Semiconductor Manufacturing Co. and other foundries already know how to do that, observed Scemama of ABN Amro. Bluetooth silicon vendor CSR plc "is already a leader in CMOS RF, and it is a fabless company," Scemama added.
The fast changes being seen in the IC industry underscore a trend toward collaboration--with partners and, sometimes, with competitors, said Jim Hines, an analyst with Gartner Inc. (Stamford, Conn.). "We've been predicting deeper levels of collaboration among semiconductor manufacturers for some time," Hines said. "This kind of collaboration is necessary. Companies must find alliances to enable them to remain successful."
At the start of this decade, a number of big chip alliances were formed to let vendors share the risks and costs of semi- conductor manufacturing and process development. One of them was Crolles2, the high-profile alliance created in 2000 in Crolles, France, by ST and Philips Semiconductors (now NXP). Freescale joined two years later, and foundry giant TSMC signed on as an associate partner.
But two weeks ago, NXP (Eindhoven, Netherlands) disclosed it will not extend its participation beyond the term expiring at the end of 2007 (see Jan. 22, page 1). Instead, NXP will expand its ties with longtime foundry partner TSMC.
What remains unclear is how the departure of TI and ST from the cockpit of advanced logic process development will affect such consortia as International Sematech (Austin, Texas) and the broad team hosted by IMEC (Leuven, Belgium). Both organizations are looking to develop technologies for the 32- and 22-nm nodes.
At last count, the IMEC 32-nm CMOS research program included 10 companies: Infineon, Intel, Micron, NXP, Panasonic, Qimonda, Samsung, ST, TI and TSMC. If half were to drop out because they lacked an abiding interest in process development, the five left to foot the bill could decide the program no longer represented compelling value and could retire to their respective corners. That could slow process development across the industry.
But Luc Van den Hove, COO and vice president of silicon process and device technology at IMEC, disputes that scenario. "The R&D still needs to be done, and that creates opportunities for a flexible R&D operation like us," he said. "A recent change was to increase the focus on memory technology, so we now have logic and memory--both flash and DRAM--in the core program."
Nor does Van den Hove see Moore's Law winding down. "Overall, and particularly in memory, there is a tremendous push to continue scaling processes," he said.
Japanese chip makers are already forging partnerships among themselves, in addition to their participation in a number of international alliances. Last year, after failing to establish a joint foundry venture, Hitachi, Renesas and Toshiba formed a three-way 45-nm technology alliance. In a separate partnership, Toshiba, Sony and NEC joined hands in 2006 to work on what appears to be a duplicate effort.
"The question mark is really against the Japanese. They tried to reorganize, but it was too little, too late," said ABN Amro's Scemama. "To go fab-lite or fabless, you have to have intellectual property."