The struggle between two radically different approaches to making insulating materials for integrated circuits has ended with a clear winner for the 90-nm generation of chips that began a full-scale rollout this year.
In September, IBM Corp. announced that it was switching from Dow Chemical's spin-on, polymer-based low- k dielectric material called SiLK to the rival small molecule chemical vapor deposition (CVD) technique. This followed a decision last spring by the world's second largest foundry, Taiwan-based United Microelectronics Corp. (UMC), to drop SiLK in favor of the small-molecule CVD Coral process from Novellus Systems Inc. (San Jose, Calif.).
”SiLK's chances are significantly damaged,” says Dean Freeman, principal analyst at Gartner Dataquest, in San Jose, Calif. ”They've lost out at the 90-nm technology node.”
SiLK's slippage is a testament to how economies of scale, not technological merit, so often dictate who wins and who loses in the manufacturing marketplace.
Since 1997, when a technology road map for semiconductors identified low- k dielectrics as crucial to keeping Moore's Law on track, the semiconductor industry has tried to lower the capacitance of wire interconnects by changing the dielectric on which the wire is deposited. Silicon dioxide, which has been used for more than 30 years, has a dielectric constant, or k value, of 4.2. And that just won't do for 90 nm, where the updated road map calls for k values between 2.6 and 3.1 in 2004 [see ”Fast Films,” IEEE Spectrum, February, pp. 36-41].
So the industry has been evaluating two different approaches to low- k materials, which currently yield k values of 2.6 to 3.0. On one side are technologies such as Coral from Novellus and Black Diamond from Applied Materials Inc. (Santa Clara, Calif.) that rely on CVD and yield k values of 2.8 to 3.0. CVD thermally cracks large molecules into smaller parts that reassemble as carbon-doped oxide thin films on the semiconductor surface. Familiar to the industry, CVD is seen as an evolutionary path to low- k chips.
The alternative, spin-on dielectrics, are composed of carbon or silicon-based polymers that are applied by pouring liquid mixtures onto rapidly spinning wafers. SiLK, from Dow Chemical Co. (Midland, Mich.), has demonstrated a k of 2.6 and could provide a path to a k below 2.2, good enough for 65 nm in 2005-2006 and perhaps even 45 nm in 2008-2010.
Getting the job done now
UMC and the world's No. 1 foundry, Taiwan Semiconductor Manufacturing Co. (TMSC, Hsinchu), are both committed to CVD at 90 nm. IBM--looking to challenge the leaders in the foundry market--has been taking on more foundry work at its own fabs.
The company seemed firmly committed to SiLK until highly publicized problems with chips it was making for Xilinx Inc. (San Jose, Calif.), maker of field-programmable gate arrays, delayed shipments of the company's high-end devices. Eventually, Xilinx opted to use fluorosilicate glass for its dielectric just to get the product out the door and, according to published reports, has since shifted the majority of its business to UMC.
”People are looking for a technological smoking gun [dooming SiLK], which they're not going to find,” says Bernard Meyerson, vice president and chief technologist at IBM's Microelectronics Division in Hopewell Junction, N.Y. ”With any new dielectric material, you face integration challenges such as differential thermal expansion, poor adhesion, and the like. We solved those problems successfully.”
But even as some IBM engineers were solving problems related to SiLK, other Big Blue engineers were working in parallel on CVD-related technology and racking up patents. Meanwhile, Agere Systems, Advanced Micro Devices, Intel, Motorola, NEC, Sony, and Toshiba, along with TSMC and UMC, were hopping on the CVD bandwagon. As more companies go with CVD, economies of scale kick in: material costs drop, as do the costs of tools. Industry consensus settled on CVD, and eventually, so did IBM.
”It came down to which technology in the long term makes the most sense from a business perspective,” Meyerson says.
Dow's spin-on process bucked up against a deeply entrenched incumbent technology and lost this round, admits Mark McClear, Dow's global business director for advanced electronic materials, who says that all of the technical challenges of integrating SiLK into 90-nm processes had been solved at Fujitsu and IBM.
”The decision goes much deeper than technology,” McClear points out. ”It gets to incumbency in the manufacturing space, and into the business model of the customer,” which in the case of IBM has shifted to a foundry model over the last year.
McClear says Dow is financially committed to seeing SiLK succeed at the 65-nm generation.
Although it might take a complete failure of CVD at 65 nm to get SiLK back in the mix, several firms like Toshiba are at least considering using SiLK as a part of a hybrid approach at 65 nm, says Gartner's Freeman.