[Dow Chemical photo]

Michael McCoy
C&EN Northeast News Bureau

Copper and low k. For the electronics industry, these words represent a whole new way of designing and manufacturing semiconductors that are smaller and faster.

For the chemical industry, the words represent a potentially huge opportunity to supply a range of chemicals and materials that will be needed to make these new semiconductors. However, they also represent a significant challenge as the shift to the new manufacturing technique will create winners and losers among materials suppliers.

For most of the semiconductor industry's short life, circuit lines that connect transistors and other chip components have been formed with aluminum metal. These thin aluminum lines--created through a multistep process of metal deposition, masking, and etching--are protected from each other with an insulating material, usually silicon dioxide.

This basic circuit structure worked well through many generations of computer chip advances--the seemingly unstoppable process of improvement known in the industry as Moore's Law. The law, named after a 1965 prediction by Intel Corp. cofounder Gordon E. Moore, accurately anticipated that the computing power of silicon chips would double every 18 to 24 months.

However, as aluminum circuit lines followed the Moore's Law curve and began to approach 0.18 m in width, the limiting factor in computer processor speed shifted from transistors--the traditional trouble spot--to the aluminum and the SiO₂ insulation material, also known as the dielectric.

According to Mark McClear, business director for Dow Chemical's semiconductor materials group, that realization precipitated a shift in semiconductor industry research from transistors to wiring. "They realized that there were two levers left to pull--one was change the metal and one was change the dielectric," he says.

In the mid-1990s, some 150 potential new dielectric materials were identified by Sematech, the semiconductor manufacturing technology consortium, whereas only three metals--silver, gold, and copper--were viewed as feasible. As a result, McClear says, most industry research focused on finding the best new dielectric.

However, in September 1997 IBM surprised the computer industry by announcing that it had decided on the metal, choosing copper as its circuit material of the future. "Suddenly, everybody's program shifted to copper research from dielectric research," McClear says. "Dielectric research was put on hold."

Jim Ryan, manager of interconnect technology for IBM Microelectronics in East Fishkill, N.Y., says his company chose the metal before the dielectric precisely because it was the easier decision. "Our strategy is to make one change, then incorporate it commercially," Ryan says. In 1998, IBM launched production of semiconductors with 0.22-m copper wiring using a conventional dielectric.

However, once IBM and other semiconductor manufacturers were able to make copper wiring work, the research focus shifted back to the dielectric. Gradually, the original 150 candidates were whittled down to just a handful.

Like it did with copper, IBM came out first with its material of choice for the new dielectric, announcing in April that it had picked Dow's SiLK aromatic hydrocarbon polymer, which has a dielectric constant--what the industry calls k value--of 2.65. Suddenly, Dow, a company with little experience in the semiconductor materials business, was linked up with one of the world's top chip makers.

Ryan acknowledges that choosing a new dielectric was difficult. "Silicon dioxide is a marvelous material," he says. "It has all the characteristics you like--no low-k material can do what it can do." But with a dielectric constant of about 4.2, SiO₂ wasn't a good enough insulator to prevent cross talk between the closely spaced wires in the smaller generation of electronic devices.

The criteria that IBM subjected its dielectric candidates to included thermal stability to 450 C, a dielectric constant of less than 3.0, good adhesion, chemical compatibility with other chip components, etchability, and commercial availability. "We looked through a whole host of dielectrics," Ryan says, "and the one we felt had the best properties was SiLK."

Dow's McClear says SiLK works so well because it was invented for this particular job. He says SiLK was developed after a Dow researcher attended a 1995 conference at which the dielectric problem was discussed. "He said, 'Hey, this is something we can invent,' " McClear says. "It isn't something we had lying around the back that we used to coat the bottom of boats."

Spin-on application of dielectrics such as Dow Chemical's SiLK aromatic hydrocarbon polymer is starting to challenge the conventional vapor deposition technique.

IBM will start producing 0.13-m semiconductors that employ copper and SiLK in the first half of 2001 at an existing plant in Burlington, Vt. A new $2.5 billion plant based on copper and SiLK--part of IBM's largest ever capital investment--will then start up in 2003 in East Fishkill.

These applications will mark the commercialization of SiLK, but McClear makes it clear that IBM won't be Dow's only customer. "There are about 15 semiconductor companies worldwide with active low-k programs, and we are working with all of them," he says.

Although dielectric materials are used in minute quantities in semiconductors, they add up to a potentially huge market. Michael Corbett, business manager at Little Falls, N.J., consulting firm Kline & Co. , predicts that world sales of low-k materials, only about $4 million this year, will grow to $300 million by 2004. McClear goes further, predicting sales will reach $500 million at maturity.

Unlike copper for wiring, however, no one dielectric will win out. McClear, for example, acknowledges that SiLK will only capture a portion of the dielectric market--he predicts 30 to 40%. And he has competitors contending that even that prediction is way too optimistic.

One of these competitors is Wilbert van den Hoek, executive vice president for integration and advanced development at Novellus Systems, a San Jose, Calif.-based maker of systems that use chemical vapor deposition (CVD) to place thin films on silicon wafers.

Equipment from Novellus and its chief competitor, Applied Materials, Santa Clara, Calif., is the mainstay in the semiconductor industry for depositing SiO₂ dielectric, which is generated in plasma form from precursors such as tetraethyl orthosilicate (TEOS) and silane. Their equipment also deposits a number of other metals and films used in semiconductor manufacture, including fluorinated silicate glass, a "transitional" dielectric--its k value is about 3.7--that is now used in a number of semiconductor plants.

Novellus' low-k solution is a CVD-applied material it calls Coral. Van den Hoek describes Coral as a silicon dioxide matrix in which some of the oxygen atoms have been replaced with methyl groups that lower the dielectric constant to 2.7. Rather than TEOS or silane, this carbon-doped silicon oxide is based on a proprietary but widely available precursor, he says.

A chief difference between Coral and SiLK is that Coral is applied using a plasma with traditional CVD equipment whereas SiLK is "spun on" to a spinning silicon wafer, dispersing across its surface through centrifugal force.

Van den Hoek admits that IBM's decision has heightened industry interest in spin-on dielectrics, but he maintains that prior to the announcement most semiconductor makers had abandoned spin-on polymers because of problems with the approach. "Those problems have not disappeared," he says.

According to van den Hoek, these problems include lower hardness compared with CVD materials; a high coefficient of thermal expansion, which makes integration into the overall semiconductor manufacturing process difficult; and a much higher price--dollars per gram versus cents per gram for CVD materials.

Of course, in semiconductor manufacture, the key term in assessing price is "cost of ownership," which also factors in equipment costs, time, manpower, product loss, and many other variables. There, too, van den Hoek maintains that spin-on application ends up costing two to three times more than CVD.

Low-k CVD dielectrics are very new: Carbon-containing silicon oxides based on trimethylsilane (TMS) were first proposed by Dow Corning researcher Mark Loboda in 1998. Van den Hoek contends that IBM chose to go with a spin-on solution because the company needed to make a decision before low-k CVD processes were established. "They had done several years of work with SiLK and came up with a process that works, although complex and high cost," he says.

Dow's McClear counters that the spin-on process, while new for dielectric deposition, is already widely in use in semiconductor plants as a way of applying photoresist--the polymeric material used in the masking and etching of circuit lines. Likewise, he asserts that cost of ownership for SiLK is "competitive with any alternative."

Moreover, because the price of SiLK will inevitably come down and less expensive products like Coral can't go much lower, the spin-on process will become even more competitive, McClear says. Add to this reductions in product loss during the spin-on process, and he claims that spin-on application will be less expensive in the long run.

While Dow and Novellus debate the relative merits of spin-on versus CVD technology, one company-- Dow Corning --is playing on both sides of the fence, according to Jim Easton, the company's global commercial leader for semiconductor materials.

On one side, Dow Corning has been successfully marketing the TMS first proposed by Laboda for use in low-k CVD deposition processes. Applied Materials, Novellus' competitor in CVD equipment, is known to base its Black Diamond low-k material on TMS, and Easton says Dow Corning plans to further expand its line of silicon-based chemicals to encompass more CVD processes.

On the other side of the fence, Dow Corning has developed its own line of spin-on dielectrics based on hydrogen silsesquioxane (HSQ). One product, FOx brand flowable oxide, has been available since 1992 and is in use in several types of computer chips. With a dielectric constant of 2.9, "it's the largest selling low-k material today," Easton says.

In July, at the Semicon trade show in San Francisco, Dow Corning announced a new product line, called XLK, that is essentially HSQ with pores. These pores, made by boiling out a solvent during a cure step in the semiconductor fabrication process, lower the dielectric constant of HSQ significantly--to 2.0, 2.2, or 2.5, depending on pore size.

Easton says a number of evaluations of XLK are active globally. "There is a lot of customer interest in XLK because, being silicon-based, it has a high degree of familiarity," he says.

Dow's McClear discounts the viability of products like XLK for the next generation of semiconductors, claiming that SiLK is "the last spin-on standing" for chips based on 0.13-m lines. Easton, however, says most customer work with XLK is on the subsequent generation--semiconductors with 0.10-m copper wiring--where a material with a lower dielectric value than SiLK will be needed.

Also playing on both sides of the fence is Honeywell Electronic Materials . But in Honeywell's case, the two sides are organic and inorganic materials--both applied by spin-on techniques. The company's product line consists of Flare, an organic spin-on polymer with a dielectric constant of 2.8; HOSP, a spin-on hybrid siloxane-organic polymer with a dielectric value of 2.5; and Nanoglass, a spin-on nanoporous silica coating with a dielectric value of 2.0.

Like Easton at Dow Corning, Michael Thomas, chief technical officer of Honeywell's wafer fabrication materials unit, says the main target for his products is the 0.10-m generation, set for commercialization in 2003 or 2004. Although some chip makers will stick with existing dielectrics at 0.13 m, Thomas says 0.10 m will be the true proving ground for low-k dielectrics because the whole industry will need to use low k at that line width.

Thomas says Honeywell's dielectric product line breadth means it can provide a "full spin-on solution," so no CVD dielectric is needed. He advises to look beyond the k value of individual dielectrics to overall, or effective, k. The CVD approach racks up a 0.5-k penalty in total dielectric value, Thomas claims, whereas "we can stack our organic and inorganic materials together so there are no penalties in the effective k."

Like the rest of the industry, Honeywell learned about IBM's choice of SiLK in April, but Lynn Forester, marketing director for Honeywell's wafer fabrication unit, cautions not to read too much into it. "It was entirely consistent with the fact that IBM is a foundry," a company that sells semiconductors on the merchant market and has reason to boast about its technical prowess. "Lots of our customers are very reluctant to disclose what they are doing technologically," Forester says.

She points to Sony's less widely noticed revelation at a technical conference in December 1999--several months before IBM's announcement--that it successfully tested Honeywell's Flare as a dielectric in chips for advanced consumer electronics.

Although Honeywell and Dow are fierce competitors in the dielectric arena, they are united in the belief that spin-on polymers are the approach of the future because they offer something customers want badly: extendability to later chip generations. "The CVD industry doesn't have any answer for the future," Thomas says.

IBM's Ryan notes that Dow and IBM are addressing the extendability of SiLK through an $18 million Advanced Technology Program grant from the National Institute of Standards & Technology . The goal of the program is to develop a porous version of SiLK with a dielectric constant of 2.0 for 0.10-m semiconductors.

Ryan says SiLK's extendability is a big part of the reason that IBM chose it. "We wanted a material that is useful for multiple generations," he says. "We will use SiLK at 0.13, and we expect to use it at 0.10."

Novellus' van den Hoek acknowledges that porosity can't be introduced into CVD materials because they aren't applied in the liquid phase. However, he says nanoscale porosity can be imparted through the use of larger organic groups--just as substitution of a methyl group for oxygen in SiO₂ gave Coral a lower dielectric value. "We have to approach the issue with a different mind-set," he says.

At the same time, van den Hoek points out that k value isn't the only necessary parameter for next-generation dielectrics. For example, he says polytetrafluoroethylene, or DuPont's Teflon, has a dielectric constant of 2.0, the lowest known among commercial materials, but it isn't suited to semiconductor processing.

In van den Hoek's opinion, the true challenge for materials companies is coming up with a dielectric that can be integrated into the overall production process--something he believes will be difficult with porous polymers. "We consider getting to lower k the least of our problems," he says. "The real challenge is coming up with a usable semiconductor manufacturing material."

While the major dielectric suppliers battle for low-k market supremacy, suppliers of other chemicals and gases used in semiconductor manufacturing are scrambling to keep up with the changes.

At Air Products & Chemicals , Joseph Stockunas, worldwide marketing manager for electronics, says his group has been able to anticipate changes in the materials needs of semiconductor fabricators because it already supplies them with some 35 gases and more than 30 chemicals. Overall, he expects the move to copper and low k will be a plus for Air Products, but he admits that the firm will take a few hits.

Today, for example, the company is a major supplier of high-purity TEOS used for SiO₂ generation in CVD equipment. As the industry moves to the first generation of new dielectrics--the transitional fluorinated silicate glass products--Air Products is ready as the dominant supplier of silicon tetrafluoride, used to dope SiO₂. The company recently completed an expansion of silicon tetrafluoride capacity at its facility in Morrisville, Pa., Stockunas adds.

For the move to the carbon-doped silicon oxides based on TMS such as Black Diamond, Air Products has struck a deal with Dow Corning to be its distributor of TMS to the electronics industry. Other low-k CVD dielectrics will be based on tetramethylsilane, and Air Products will supply this chemical through its Schumacher subsidiary, he adds.

On the down side, Stockunas notes that Air Products sells chlorine-based gases such as boron trichloride that are used to etch aluminum; however, copper isn't etched, so sales of these products--already relatively small volume--will eventually decline. Likewise, Air Products is the leading producer of tungsten hexafluoride, which is used to create tungsten interconnects to aluminum wiring. But the new chips will use copper for the wiring and the interconnect, so tungsten hexafluoride volumes will eventually fall as well.

Stockunas and other industry observers emphasize that, initially at least, copper/low-k computer chips will be niche products aimed at the high end of the electronics market. They point out that the traditional aluminum-SiO₂ semiconductors will be the dominant product for years to come and, in fact, will continue to grow, spurring chemical demand along with them.

For some chemicals, such as nitrogen trifluoride, the emergence of new dielectrics may mean a slight trimming of the growth curve. Currently, NF₃ is experiencing soaring demand as a CVD chamber cleaning gas. These chambers are used, among other things, to deposit dielectric, but as the industry transitions to low k, some of the dielectric will be spun-on without the use of CVD chambers.

Air Products, the world's largest NF₃ producer, is in the midst of another expansion at its Hometown, Pa., plant in response to a worldwide shortage of the chemical. Stockunas acknowledges that spin-on application of dielectric will eventually affect NF₃ demand, but for the most part he doesn't see the technique really emerging until 0.10-m chips come along several years from now. "The NF₃ market is very tight right now," he says. "We're putting the capital in the ground to meet the market's needs for next year."

In addition to NF₃ and other chemicals closely tied to dielectric production, ancillary products such as photoresist strippers and plasma etch residue removers will also be affected by the move to copper and low-k dielectrics.

Ashland Specialty Chemicals is a major supplier of these strippers and removers through the Ashland-ACT unit of its electronic chemicals division. And Bob Rohlfing, the unit's director, says product development related to copper and low-k dielectrics is advancing quickly. "It's a major activity that's deciding how we allocate resources," he says. "Right now, it's disproportionate to actual sales figures."

Many of the strippers and residue removers used with aluminum-based semiconductors contain hydroxylamine, a product that is in short supply because of an explosion at a plant in Japan owned by Nisshin Chemical, Ashland's main supplier ( C&EN, July 24, page 23 ). However, while stripper makers are scrambling for hydroxylamine today, it isn't compatible with copper, so new chemistries will have to be developed for chip fabrication steps in which copper is involved. So far, Rohlfing says, Ashland has delivered two products--the NE series and ACT 970--that work well with copper.

At the same time, producers of strippers and residue removers must make sure their products interact well with the new dielectrics. Ashland has been informed that its NE-14 is the process of record for the Coral and Black Diamond CVD dielectrics, Rohlfing notes. Likewise, the company is working with Dow on ensuring the compatibility of its etch residue removers with SiLK.

Rohlfing says the ancillary chemicals market is fracturing because there's no standardization of the copper/low-k chip materials like there is with aluminum and SiO₂. "Things are becoming much more customer specific," he says. "I believe we will have some high-volume products like the NE series and ACT 970 in the future that work over a range of applications. At the same time, we will have a large number of customer- and application-specific products that are only as successful as the customer is."

The fast-growing field of chemical mechanical planarization (CMP)--the use of dilute abrasive slurries to polish, or planarize, silicon wafer surfaces--will also change as a result of the move to copper and low-k dielectrics. Today, the CMP business is dominated by Cabot Microelectronics, Aurora, Ill., with an estimated 80% market share, but a number of firms see the shift to copper as an opening into the market.

One of these is Arch Chemicals , which recently formed Planar Solutions, a joint venture with Wacker Silicones to develop and market CMP slurries. Philippe Gouby, vice president and general manager of Arch's microelectronics unit, says the joint venture, finalized in April, already has several products on the market.

However, Gouby admits that Planar Solutions is something of a latecomer to the mainstay CMP market--the polishing of aluminum and SiO₂. "We have a very good product, but because of the dominance of Cabot in this field, we have also come in with a high-performance copper slurry," he says.

Ashland is taking a similar approach. Charles W. Cook Jr., general manager of its electronic chemicals division, says Ashland chemists have developed CMP slurries targeted at copper, as well as specialized post-CMP cleaning compounds. "We have products in the latest stages of development," he says. "If we can demonstrate our capability in copper, then we can back into other applications as well."

Bayer Corp. is getting into CMP too, according to Elie Saad, director for electronic chemicals, with a focus on copper and low-k dielectric polishing. Bayer is offering the standard CMP products, but Saad thinks the company's best chances are in the emerging markets where no company has an established product line. "Chip fabricators have to test before they jump into copper," he says. "Bayer will have just as good an opportunity as anyone else."

Existing CMP suppliers caution that making it in CMP, even in the new copper segment, won't be easy. "There's much more to it than putting chemicals together and sending them out to someone," says Don Frey, director of marketing for CMP at EKC Technologies , a Hayward, Calif.-based unit of ChemFirst that is one of the top three CMP companies.

Copper itself is a new market for chemical suppliers. In July, for example, Arch Chemicals linked up with MacDermid Inc. to develop and sell copper electroplating systems for the new semiconductors. Ed Wentworth, business manager for Arch's thin-film systems unit, says Arch is also developing a CVD-applied copper "seed" layer to be put down before the electroplating step.

In today's chips, a thin film of titanium nitride is applied via CVD before the aluminum is deposited, but Arch is currently evaluating numerous replacement candidates because, according to Wentworth, it's not clear that the current approach will work with copper. "We're not just optimizing the same old materials," he says. "We're playing with brand-new molecules."

Clariant's AZ Electronic Materials unit in Somerville, N.J., envisions an expanded opportunity for its line of bottom antireflective coatings used to protect the metal as well as other circuit components during various etching steps. According to Product Manager Kathryn Yager, AZ had to develop an entirely new line of coatings that work with copper, but because of the expanded need for them in the copper circuit architecture, the payoff should be a new market twice as big as the existing one.

Kline & Co.'s Corbett sees all this activity as good news for chemical suppliers, who will have the opportunity to play a partnerlike role in the more varied materials landscape of the future. "Today, semiconductor makers beat the heck out of their suppliers," he says. "But in the future, they will need their chemicals to be customized to a specific application."

At Ashland, Cook agrees that chemical suppliers are doing a lot to make the new copper and low-k technology work for semiconductor producers, but he's not convinced that the industry will be in a stronger negotiating position as the new materials proliferate. "I think the value of what we do will be easier to see," he says, "but whether or not we can retain that value remains to be seen."

[Previous Story] [Next Story]

Top

IBM is first to use copper and low-k dielectric in semiconductors

IBM will use copper wiring and Dow Chemical's low dielectric constant SiLK polymer in a new generation of semiconductors to debut next year. [Image courtesy of IBM]

[Previous Story] [Next Story]

Top

Of the major electronic chemicals segments, the one least affected by the move to copper circuitry and materials with low dielectric constants is high-purity wet chemicals used for semiconductor wafer cleaning and preparation. However, this business has more than enough problems of its own to contend with.

According to figures released midyear by Semiconductor Equipment & Materials International, a San Jose, Calif.-based trade association, the global semiconductor industry is expected to grow 30% this year to $195 billion in sales. But sales of acids, bases, and solvents used in semiconductor production will rise only 10% to about $745 million.

Most chemical businesses would kill for such double-digit gains, but wet chemicals makers embarked on a major capacity buildup in recent years, anticipating even higher growth. The most recent victim of the muted outlook is Arch Chemicals, which three weeks ago announced a major downsizing of its wet chemicals business, which it calls process chemicals.

Arch will take a $50 million charge to cover the cost of cutting back manufacturing in Mesa, Ariz., and Zwijndrecht, Belgium; downsizing product offerings; and refocusing on formulated specialty chemicals. In the process, it expects to reduce sales by more than 75% but improve operating margins to 10% of sales. Arch's process chemicals business reported a loss of $1.8 million in the third quarter on sales of $17 million.

Philippe Gouby, vice president and general manager of Arch's microelectronics unit, says the company spent $75 million over five years to build up the process chemicals business, but it got caught by the triple forces of the big capacity buildup, a sharp semiconductor industry downturn in 1996-97, and the development of new cleaning techniques that cut demand for its products.

As a result, he says, Arch will exit "straight" chemicals--basic high-purity products to which Arch doesn't add much value--and focus on specialties such as etchants, strippers, and residue removers for which the company offers special know-how. "These products will be really important, even strategic, to our customers," Gouby says.

Arch's downsizing follows several other restructuring moves in wet chemicals. Early this year, Houston-based Koch Microelectronic Services, a newcomer to the field, put its business up for sale. Then Honeywell Specialty Chemicals and Mitsubishi Chemical announced in July that they intend to combine their wet chemicals businesses into a new jointly owned company. The deal is expected to be final by year's end. Honeywell and Mitsubishi expect their alliance to be number two in wet chemicals sales worldwide, after Ashland Specialty Chemical's electronic chemicals division.

At the Ashland division, General Manager Charles W. Cook Jr. says he has been trying to ignore the upheaval and stay the course. For Ashland, this meant starting up a new wet chemicals plant in 1998 in Pueblo, Colo., right in the midst of the downturn. A similar plant set for Taiwan was delayed about a year, but started up last month.

Cook says the Taiwan plant is coming onstream in a wet chemicals industry environment that has improved modestly this year along with rising semiconductor production. However, he anticipates that the business is on the cusp of a real step change in demand as new semiconductor fabrication equipment built in recent years finally comes on-line, requiring healthy initial chemical doses. "The meat of the recovery will happen as new capital equipment comes onstream," Cook says. "That's just now beginning."

[Previous Story] [Next Story]

Top

Chemical & Engineering News
Copyright © 2000 American Chemical Society