The lithography equipment marketplace
01/01/2001
Robert N. Castellano, The Information Network, New Tripoli, Pennsylvania
Several factors shape the lithography equipment sector compared to the total equipment market, including the huge cost of the equipment, the volatility of tool sales, the major players, and the extendability/limits of current optical technology. Accurate interpretations of market data and future predictions depend on an understanding of these changing forces.
Spiraling costs
Lithography is the single largest cost factor in semiconductor production. According to a rule of thumb developed by The Information Network, exposure systems make up 20% of the total fab cost. When the price tag for a new fab hits $10 billion sometime in this decade, the lithography bay alone could account for $2 billion of that cost — or more than what the entire fab costs today.
What's more, lithography costs have risen faster than overall costs. It used to take about 10 exposures to make a 1Mb chip; now 64Mb units take 23 mask layers, and 256Mb units require 27. In addition, critical resolution continues to drop. While 16Mb DRAMs demand 0.5—m rules, 256Mb chips must have 0.25—m criticality.
Worldwide R&D costs for lithography tools reached $538 million in 1999, and were primarily for 193nm tools. R&D costs for photomasks reached $143 million in the same year with the bulk of these funds going toward phase-shift masks (PSMs) and optical proximity correction (OPC) masks. R&D costs for photoresists reached $116 million in 1999, primarily for 193nm and advanced 248nm resist.
 Figure 1. Lithography market compared to the total equipment market. |
In addition, millions of dollars were spent in 1999 on next-generation lithography (NGL) tools such as extreme ultraviolet lithography (EUV); x-ray lithography (XRL); electron projection lithography (EPL); and ion projection lithography (IPL). EPL technology has been approached in three different ways: SCALPEL favored by Lucent, ASML, and Applied Materials; PREVAIL from IBM and Nikon; and an e-beam based on field-emission display (FED) technology from Motorola. Many of these highly complex systems are years away from a production environment. SCALPEL, for example, is only expected to be ready by 2004-2005, and to have a 15wph throughput. COO for this system is high, but one can only guess at the mean time between failure rate for the FED system, a $15 million, four-chamber cluster tool with two e-beam writing chambers, and a whopping 201 columns, each with 32 field emitters.
The Information Network estimates that $400-500 million/year of private and government R&D money is pouring into NGL schemes. Based on industry workshops conducted by International Sematech, the total R&D spending for one regional solution could approach $1 billion in 2002 alone, almost double previous spending rates.
Volatile tool sales
Figure 1 compares the scope of the lithography market to the total equipment market, and shows how the percentage of lithography equipment sold has varied over the past decade from a low of 7.1% in 1993 to a high of 15.1% in 1999.
The step-and-repeat aligner sector is the most volatile. Sales dropped 35.8% in 1986 from $413 million in 1985 to $265 million in 1986, rising to $355 million in 1987, $1.0 billion in 1989, and $1.3 billion in 1990. Revenues dropped 4.7% in 1991 to $1.2 billion and 25.0% in 1992 to $917 million. Revenues dropped 28.3% from $4.3 billion in 1997 to $3.1 billion in 1998. Revenues in 1999 reached $3.7 billion, and are projected to grow to $8.4 billion in 2002.
Step-and-scan tools typically offer better CD control across the field and smaller lens placement errors compared to step-and-repeat tools. Linewidth uniformity across the chip is approximately a factor of two better in a step-and-scan system than in a step-and-repeat. However, the purchasing decision is not black and white. For example, Nikon sold 107 248nm step-and-scan tools and 35 step-and-repeat tools in 1999, while ASML sold 109 248nm scanners and 12 steppers.
In 1999, 248nm DUV tools dominated the lithography market with a 57.2% share, while 193nm DUV tools held only a 2.8% share. The push to finer feature sizes in the semiconductor industry will drive the DUV market.
Major players
Nikon has led the lithography marketplace for more than a decade, as shown in Fig. 2, and did so in 2000 despite the pending acquisition of Silicon Valley Group Lithography (SVGL) by ASM Lithography (ASML) announced last October. (See "The ASML-SVGL merger: Refractive versus catadioptric lenses" below for information on the technology behind the merger.)
 Figure 2. Lithography market shares. |
In terms of number of tools sold in 1999, Nikon was the market leader with 270. Combining ASML's 195 units and SVGL's 50 units, Nikon remained the leader through the end of 1999. ASML had 137 and SVGL had 50 DUV shipments for 1999, putting the combined companies' total ahead of Nikon with 150 DUV tools sold.
Exchange rates played a key factor in determining revenues. ASML's 1.2 billion euros equaled US$1.3 billion at an exchange rate of US$1.0668/euro at the end of 1999, but only $1.1 billion at the exchange rate of US$0.8786/euro at the time of the acquisition announcement. Adding SVGL's CY revenues of $236 million brings 1999 numbers to either $1.5 billion or $1.3 billion for the combined companies. Nikon had 1999 revenues of $1.4 billion.
Optical litho
The extendability limit of optical lithography continues to be pushed out. In 1991, for example, optical lithography was expected to end at the 180nm node with 248nm DUV lithography. The 193nm DUV era was originally set for the 180nm technology node. At the 130nm node, it was assumed the industry would move to nonoptical NGL. Many in the semiconductor industry believe 157nm optical lithography will be introduced at the 0.07—m node.
COO and physics may play a role in determining the outcome of the lithography technology race. ICs with 130nm features have already arrived, and if the density of ICs continues to climb according to Moore's Law, then feature sizes should dip to the 100nm node by the year 2005.
Through optical extension developments, 248nm lithography has not only secured the 150nm slot, but it is also expected to support CD requirements down to the 130nm node. Optical extensions — PSMs and OPC — can extend the subwavelength range of optical lithography into the deep submicron realm. They have been used in R&D labs since the early 1980s, but it wasn't until the 250nm node that these methods were needed, because advances in semiconductor technologies had been met with advances in lithographic equipment.
 |
Figure 3. Lithography requirements and exposure tool potential solutions. The legend indicates the time at which research development and qualification/pre-production should be taking place for each solution. Note: Production-level exposure tools should be available one year before first IC shipment.
Optical lithography continues to push toward resolution limits undreamed of a decade ago. ASML, for example, believes that the limit of optical lithography lies somewhere between 70nm and 50nm feature sizes. The company finds that it is even conceivable that optical will succeed in producing details of 35nm with lenses and photons.
The results of several programs have also suggested that this could be accomplished:
- International Sematech has been producing 130nm dense features with good process latitude and depth of focus on a 193nm full-field system. Scientists there are creating 100nm gates with binary plates and 70nm gates with phase-shift technology.
- MIT's Lincoln Labs, using NumeriTech's phase-shifting technology and 248nm DUV stepper, fabricated a functioning CMOS transistor with 0.025—m gates.
- Silicon Valley Group Inc.'s lithography subsidiary says its Micrascan-V 193nm scanner will be able to pattern 0.07—m isolated lines and 0.10—m nested lines without the need for expensive PSMs.
- ASML announced that nine chipmakers have joined its consortium to deliver 157nm-wavelength exposure tools by 2003. Infineon Technologies AG, Motorola Inc., Philips Semiconductors, Taiwan Semiconductor Manufacturing Co., and STMicroelectronics are among the companies that have joined ASML in the 11-month-old program. ASML hopes to create a 300mm-capable, 157nm scanning system that will exceed the requirements for 70nm design rules spelled out in the International Technology Roadmap for Semiconductors (ITRS). Compared to NGL techniques, 157nm has highly attractive characteristics. The most important is that exposure at atmospheric pressure is possible, whereas vacuum is required for all NGL techniques. The greatest challenge in exposure with 157nm radiation is keeping the beam free of oxygen, since this gas is not transparent to 157nm.
In fact, the 1999 version of the ITRS has pushed out NGL from the 130nm node to 100nm and below compared to the 1997 version. Technology options from the 1999 ITRS are shown in Fig. 3. (Also, see "Nodal predictions" below for future technology options at each node.)
The ITRS Lithography Technology Working Group has also put more pressure on maskmakers to address mask error function and PSM requirements, which are critical to extending a tool at least one generation.
NGL
How far can optical lithography extend? Could it eventually make NGL technology unnecessary?
There are no fewer than four uncertain NGL technologies from which to choose: EPL, proximity XRL, IPL, EUV, and e-beam direct write (EBDW) for use at the 70nm node as shown in Fig. 3 on p. 62. It is estimated that each of these new lithographic methods will require a $500 million investment. Which technology wins out will depend not only on the system but also on the COO of the tool itself and the masks.
A huge infrastructure of worldwide equipment and semiconductor manufacturers is developing the NGL alternatives. One of the most active areas in terms of programs and backers is EUV, which uses a 13nm wavelength pulsed xenon gas laser and has the capability of operating below 0.1—m linewidth. Military and defense company TRW's system, for example, uses a laser interacting with a jet of xenon gas to produce light at 13.5nm. It was developed as a joint venture with EUV LLC, a US-based consortium that includes Intel, AMD, Infineon, Micron Technology, Motorola, ASML, SVGL, and Virtual National Laboratory, which combines resources from the Lawrence Livermore National Laboratory, Lawrence Berkeley Laboratory, and the Sandia National Laboratories. A European initiative includes ASML, Zeiss, Oxford, and Infineon, while a Japanese program includes Nikon and Canon.
Complicating the scenario is the fact that an EUV system uses mirrors to achieve resolution down to 25nm that, along with the mask, all have 80 layers of a coating material of molybdenum and silicon, 1000 atoms thick. The Information Network maintains, however, that EUV will win out as the NGL of choice, but not until the 50nm node, as optical will continue to push the envelope.
Robert N. Castellano is president of The Information Network, a consulting and market research company addressing the semiconductor, computer, and telecommunications industries. This article is based on the report Sub 0.25-Micron Lithography: Market Analysis and Strategic Issues. For further information, contact the author at 8740 Lyon Valley Rd., New Tripoli, PA 18066; ph 610/285-4548, fax 610/285-4547, e-mail www.theinformationnet.com.
The ASML-SVGL merger: Refractive versus catadioptric lenses
Companies base their lithography purchases on specifications, willingness to agree to meet customer requirements on the wafer, COO (projected), support capability, and relationship history (either direct or from colleagues).
The performance (on paper) of all recently announced products is very close with respect to speed, resolution, alignment, light source, and even price. The major difference between SVGL and Nikon, ASML, and Canon, is that SVGL uses a reflective system with catadioptric lenses and the others use refractive lenses.
 |
Catadioptric versus refractive optic lens costs. SVGL's catadioptric Micrascan-V uses less CaF2 and is thus less expensive than refractive optic lenses
At issue is whether refractive lenses are able to function in a 157nm DUV tool or whether they are limited to chromatic aberration. If the latter is the case, then the SVGL patents on catadioptric lenses could be an advantage for ASML when the 70nm node is reached in 2005. This extendibility issue may be a significant factor in why Intel, Samsung, and Hyundai funded the development of SVGL's 193nm DUV step-and-scan tool.
Even for 193nm tools, the refractive lens technology will be a COO issue. According to SVGL, at 193nm, its Micrascan-V needs only 2kg of calcium fluoride (CaF2) for its lens, compared with the 50kg needed for refractive optics. Priced at $10,000/kg of CaF2, the materials cost is only $20,000 for the Micrascan-V lens, compared with $500,000 for the refractive optic lenses (see figure).
For 157nm DUV lithography, CaF2 material will be the only viable lens material for both refractive and reflective designs. Quartz used in current DUV tools does not have sufficiently high transmittance at 157nm, while other materials have significant limitations. For example, the birefringence of MgF2 is too high and LiF has a solubility 160 times greater than CaF2.
BaF2 is under evaluation as a lens material at 157nm. It could solve the chromatic aberration problem encountered with CaF2 when using refractive step-and-scan tools and eliminate any patent problems for Canon and Nikon. However, it too has a high solubility in water, making polishing difficult and moisture a problem.
Nodal predictions
At the 130nm node, we see a 193nm DUV tool or 248nm DUV with optical extensions. At the 100nm node, one has greater choices: 157nm DUV, 193nm DUV with optical extensions, and a plethora of NGL tools —EPL, EBDW, XRL, IP, and EUV. At the time frame for the 100nm node, which should be 2002 or 2003 for development work, NGL tools will not be production ready, so the choice will remain a DUV optical one.
At the 70nm node, 157nm DUV with optical extensions or any of the NGL tools could be used. We see 157nm DUV as the choice in the 2005 time frame for development work, even though NGL tools will be production ready. Semiconductor manufacturers will use optical extensions on 157nm tools, technology they have been using for several years and for which they have mastered the learning curve.
The time frame for the 50nm node will be 2008 for development work. Although NGL tools are expected to be the only viable technologies available, we see continued advances in DUV processing with respect to tools and materials, and we believe that DUV will remain the technology of choice among semiconductor manufacturers.
When the time comes for NGL tools, which we see happening at the start of the next decade at the 35nm node, then the choice will be between EUV and EPL. According to International Sematech, EUV critical issues remaining to be addressed include defect-free multilayer coated mask blanks; defect-free mask manufacturing; source and condenser optics reliability; COO; and reticle defect control solution.
For EPL, the remaining critical issues include demonstration of seam blending/stitching; defect-free mask manufacturing, including stress control for membrane and stencil masks; experimental data on beam blur and its relation to throughput versus feature size; reticle defect control solution; and demonstration of a solution for wafer heating. EUV and EPL manufacturers have 10 years to work on these critical issues. Because of the heavy R&D expenses involved, only those with deep pockets will continue to invest the time and money.