HALF MOON BAY, Calif. – Despite a sudden rise in capital spending, the IC equipment industry possibly faces another dreaded downturn in 2011.
The probable slowdown is expected to accelerate the ongoing shakeout in the fab tool industry, leaving some to wonder about the overall health of the business. Some believe the fab tool business model is simply broken. Over the years, fewer fabs have been built, causing massive consolidation in the equipment sector.
Fewer choices in the equipment vendor base could cause higher tool prices, less innovation and monopolies in some tool segments, such as ion implantation, lithography and metrology. Still others believe that fab tool vendors must consolidate and pool their R&D resources to tackle the next-generation technology nodes, the extreme ultraviolet (EUV) lithography supply chain, and the potential shift to 450-mm fabs.
Two big fab tool buyers-Intel Corp. and GlobalFoundries Inc.-have markedly different views about the subject.
Consolidation in the industry has put the fab tool supply chain in peril, because there are ''not enough choices’’ in the various segments in the vendor base, said Norm Armour, vice president and general manager for the Fab 8 project in New York state for GlobalFoundries, the foundry spinoff of Advanced Micro Devices Inc. (AMD).
Robert Bruck, vice president of the Technology and Manufacturing Group at Intel, has a different viewpoint. ''Last year at (a SEMI event), Bob Bruck posed the hypothesis to me that consolidation (in the fab tool sector) was good for customers,’’ wrote VLSI Research CEO G. Dan Hutcheson, in a recent report.
''Being an economist, firmly grounded in anti-trust teachings, my gut pushed back. But Bob came up with a good point: fewer competitors means less duplicative R&D, which makes the equipment industry more efficient, which should translate into lower cost,’’ Hutcheson wrote.
In any case, the fab tool industry has undergone a sea of change. During the early days of the IC industry, chip makers built their own fab tools, because there was little or no semiconductor equipment business to speak of.
Then, starting in the 1950s and 1960s, a new crop of independent semiconductor equipment maker burst onto the scene. Chip makers began to buy tools from a plethora of new startups in the arena. The fab tool business prospered in the 1980s. In 1981, there were over 300 vendors that sold equipment to 300 or fewer chip makers, many of which had wafer fabs, according to VLSI Research.
At the time, there were multiple vendors in each fab tool segment, giving chip makers a viable choice. For example, in the mid-1980s, the lithography market included ASAT, ASML, Canon, Eaton, GCA, Nikon and Ultratech. Another firm, Micronix, was developing an X-ray stepper. IBM and Japan Inc. were separately devising large-scale, synchrotron X-ray systems for next-generation devices.
In the same decade, the fab tool business model began to fall apart. In 1987, pure-play foundry Taiwan Semiconductor Manufacturing Co. (TSMC) emerged, propelling the fabless industry. Since the emergence of foundries, this model has not only propelled a new crop of fabless chip makers, but it also forced integrated device manufacturers (IDMs) to re-think their fab strategies.
Over time, a growing number of IDMs moved towards the fab-lite model. In 2001, for example, there were 16 logic IDMs that had fabs. ''Intel, Samsung, and potentially STMicroelectronics will be only IDMs with production wafer capacity at 22-/20-nm in 2012/2013,’’ said Handel Jones, an analyst with International Business Strategies Inc.
Over time, only three foundries-TSMC, Samsung, and GlobalFoundries-will have the resources to build new fabs and keep up with Moore’s Law. In memory, Samsung and Toshiba will survive over the long haul. Beyond that, there are questions in the current memory landscape.
Fewer fabs, coupled with the cyclical downturns, have lead to considerable consolidation in the fab tool arena. In 2011, there are estimated to be some 140 tool vendors that sell equipment to only 20 customers with fabs, according to VLSI Research.