Charting different roadmaps for lithography at Japanese litho firms
11/01/2001
Griff Resor, Resor Associates, Boxborough, Massachusetts
 |
overview
As an industry, we expect optical lithography to run out of capability any day, so we want a replacement technology to be ready just in time. This has been going on for nearly 25 years. Some people think the demise of optics is only a fear, that we will never need a replacement technology. History is on their side. This view supports the "Optics Forever" roadmap. But few major IC makers are willing to bet their entire future on this outlook. So the quest for a viable "next-generation lithography" technology continues, at least in Japan.
The technical alternatives have not changed substantially since 1978, when the US launched its very high-speed integrated circuit (VHSIC) program. At that time, optics was challenged to meet a 1.0µm resolution goal. E-beam direct write was selected as the replacement technology and was asked to meet a 0.5µm resolution goal.
X-ray proximity printing and ion projection lithography soon entered the competition as technical alternatives. For the past 23 years, e-beam direct write (aka electron projection lithography [EPL]), x-ray proximity printing, and ion projection lithography have remained the main contenders to serve as the industry's next-generation lithography (NGL) technology. Some fear the label "next" may be all too accurate, implying there will never be a "now."
 |
Champions of each NGL technology have competed vigorously. Since 1986, Sematech has funded NGL alternatives so that a replacement technology would be assured. Performance goals for resolution, overlay, and timing were established and published in the National Technology Roadmap for Semiconductors (NTRS). This has been updated regularly as optical improvements have forced the goals to change. The expectation has been that one NGL would emerge as the best choice, and funding of the others could be dropped. Reduction of the alternatives is now underway, stimulating some diversity of opinion as one might expect.
The optics forever roadmap
Despite the diversity of options, optical lithography has persisted. Shorter wavelengths, faster machines, DUV resists, and mask-based resolution enhancement techniques (RET) such as phase shifting have provided the continuous improvement needed. The table shows the present optics forever roadmap. Tools operating at 248nm wavelengths are today's workhorses, printing 180nm features in production. Optical processes will shift to 193nm tools in the next few years. Tools operating at 157nm should follow after that. Finally, extreme ultraviolet (EUV) tools complete the picture near the end of this decade.
The magenta boxes in the table highlight the expected evolutionary path for optical technology. For each wavelength, a learning curve permits better lenses (higher numerical aperture [NA]) and more aggressive processes (lower k1). An overlap with the next wavelength is planned to permit smooth transitions. While not shown, each wavelength generation will also persist longer in time, being used for less critical layers in a mix-and-match mode with the newer-generation tools. Today's basic architecture of step-and-scan motions is expected to persist. Very high throughput between 120 and 150 wafer levels/hr (wlph) on 300mm wafers is expected. NGL technologies must compete against this optics forever roadmap.
Intel has quite openly started to push for a winnowing of the choices. X-ray proximity technology has been dropped by nearly all IC chipmakers, except for those with specialized needs. In November 2000, e-Lith abandoned its support of SCALPEL, derived from Lucent's direct write e-beam technology. In the Intel-led club, EUV has emerged as the winner.
Not everyone agrees with these choices. In particular, the Japanese suppliers of lithography tools are taking a significantly different approach to NGL. These choices are most likely market-driven. This could be the first real sign that NGL may soon be the now-generation lithography.
Nikon tries to prevail
Nikon continues to fund the full optical roadmap. The company says it is working hard to remain technology neutral, so it has major R&D programs for e-beam, 157nm, and EUV. But there are significant differences in emphasis. For example, at the FSI breakfast during Semicon West last July, Nikon said that its customers want even higher NA lenses for use at the 248nm wavelength, to extend the resolution capability of 248-nm processes. Nikon's plan to extend 248 also includes machine optimization for phase shifting masks and double exposures required by some phase shifting technologies. Lithography at 193nm did not get the emphasis that others have provided, even though Nikon has competitive plans and higher numerical lenses in its pipeline for 193nm.
 Figure 1. Left: Nikon's PREVAIL alpha system being integrated. Right: PREVAIL imaging concept of curvilinear variable axis lenses (CVAL). |
After 193nm, Nikon's roadmap changes significantly. While Nikon is working on both 157nm and e-beam direct write systems to meet the 70nm node, the company now expects the PREVAIL e-beam direct write system will be ready first. Nikon reports that some customers are concerned that 157nm technology may be late and they want an alternative. Nikon has funded a joint program with IBM to bring IBM's novel PREVAIL technology to market. The goal is to catch the 70nm node for contact holes and gates in logic devices in CY04.
PREVAIL uses a 4x-stencil mask. Subfields 1.0mm2 are projected onto the wafer using a 4x-reduction electron lens. Approximately 10 million resolved features (pixels) can be printed in one exposure. The illumination beam of electrons fills the full 1.0mm mask subfield and is deflected to each subfield at high speed. At the same time, the wafer is moved so that full chip patterns are assembled from subfield images. A throughput of 30 wafers (200mm) is expected.
PREVAIL uses a novel reduction lens technology perfected at IBM. The optical axis of the 4x-reduction lens moves to follow the illumination scan across a 26mm-long stripe of 1.0mm subfields on the mask. This technique minimizes off-axis-imaging problems that most e-beam direct write systems have encountered. The result is better resolution and better stitching of subfield boundaries as the full wafer pattern is exposed.
The proof-of-concept column was completed at IBM in 1998. The first full system is being integrated in Japan, as shown in Fig. 1. Figure 2 shows 80nm contacts achieved with the PREVAIL tool.
Nikon expects to have the first introductory units available in CY03. Production units are planned for CY04. As Nikon has become more familiar with the PREVAIL technology, it has accelerated its schedule by more than one year. The company has said this new schedule is required to meet the continued rapid pace of shrinking geometry. It is possible that the schedule changes have been made to meet specific customer needs.
At the FSI breakfast, Nikon announced that it would have introductory units for 157nm available in CY04, and will ramp to volume production by CY07. This is about one to two years later than Canon's announced schedule, and two to three years later than Silicon Valley Group's (now ASML) plan for 157nm tools.
Nikon is active in Japan's EUV program and has said it is doing lens design studies at this time. Present plans call for beta units by 2005 and production units in 2007. It is possible that Nikon is waiting for some resolution of key technical risks in EUV technology, such as finding an adequate source, before moving to a production design.
Canon takes a shot at 157nm
Canon's optical roadmap is closer to the optics forever plan. Most recently, the company has emphasized its 193nm products. These can be used in mix-and-match mode with 248nm and i-line lenses on the same high-throughput platform for 300mm wafers. At the 157nm wavelength, Canon is planning to have introductory tools available in CY03, with production units available in CY05. This falls between ASML's and Nikon's plans.
 Figure 2. 80nm contacts from PREVAIL. |
The push to bring 157nm tools to the market early appears to be real and immediate. In July, Canon and Infineon announced a joint development program to put Canon's 157nm machines into production as quickly as possible. Infineon will conduct process development work jointly with Canon at Canon's facility in Utsunomiya, Japan, starting in 2QCY03. By the end of CY04, Infineon expects the tool and its process to be up and running at Infineon. Canon has made a major investment in CaF2 production, to assure an adequate supply for its 193nm and 157nm ramp to volume production.
Canon's roadmap departs from the optics forever roadmap after 157nm. Canon puts a novel e-beam tool, its ML2 product, next in its batting order. A CY05 introduction is planned, with initial production units available in CY06-07.
Canon is not planning to use e-beam technology to replace optics on volume production lines. The company sees a significant business opportunity for its novel e-beam approach. Canon sees a market segment developing for maskless production tools. For many chips, the production run is relatively short and mask costs dominate. The same is true for R&D cycles during chip development. In both cases, manufacturers and engineers would like to pattern wafers directly from CAD (computer aided design) data, eliminating the mask fabrication, inspection, and repair cycles and those costs. Canon figures that a maskless e-beam tool will be cost-effective in these applications, even if it is slower than high-volume production optical tools.
The key to Canon's system is a multichannel illuminator. A single source is imaged by an array of 64 x 64 miniature electrostatic lenses to form 4096 beams of electrons. These are controlled individually to write with 4096 beams at once.
The basic arrangement of the Canon system is shown in Fig. 3. Two electrostatic lens arrays labeled Lens 1 and Lens 2 create the array of 4096 beams. Each beam may be blanked on and off individually. The array of 4096 spots is placed nominally at the object plane of the main reduction lens. A reduction of 50x is used to enable practical fabrication of the lens array in the illuminator. The 12.8mm2 area illuminated by the microlens array is reduced to a 256µm2 subfield at the wafer. Electron scanning of only 4µm in the main projection lens is used to paint in all µ-fields and subfields simultaneously. Continuous stage scanning is used to move the 4096 beam "paint brush" across the wafer until the full wafer pattern has been made.
Each beam forms a 25nm spot at the wafer. Four spots are used to form a minimum feature. A four-pass grayscale writing strategy can be used to provide fine CD control and feature size tuning.
The Canon lens array performs two more unique functions in this architecture. If you look carefully at Fig. 3, you can see that each beam is focused to a different height in the nominal object plane (labeled "image of source"). This is used to correct for focal plane curvature in the main projection lens.
The array of apertures at the top of the Canon assembly includes small electrostatic deflection capability in x and y. This is used to compensate for image placement errors in the main projection lens. Both static correction and dynamic correction are provided.
These last two features give the Canon architecture its name, "Correction Lens Array" or CLA. In conventional e-beam systems, where one beam paints in the full field, dynamic corrections are applied in the main projection lens to flatten the scanned field at the wafer, and to apply scan deflections to correct for distortions. In the Canon system, with 4096 beams writing at once, such a single-beam correction system is not practical. So the microlens array must make these corrections. By applying small corrections for focus and position to each beam in the illumination half of the column, and by limiting the main projection lens scan distance to just 4µm, it seems reasonable that Canon will achieve very tight image placement tolerances and excellent field stitching.
Canon expects throughput will be 56wlph (200mm) using a 3µC/cm2 resist. This calculation is for a 100nm resolution goal. When the resolution goal is moved to 70nm, initial throughput falls to 34wlph (200mm). Improvements in resist contrast are expected to move throughput at 70nm resolution back to 52wlph (at 200mm).
Even if Canon does not reach the calculated throughput, the maskless operation should be its strongest feature. Mask costs for EUV systems are expected to be more than $1 million/set. Eliminating just 15-20 mask sets will pay for the full cost of a Canon Multi-EB Direct Write system. When a value is placed on time to market, the maskless feature becomes even more important.
 Figure 3. Canon's maskless ML2 multi-E-Beam system. Each beam is focused to a different height in the nominal object plane (labeled "image of source"). |
Canon's plans for EUV call for a parallel introduction with the ML2 e-beam machines. EUV introductory units are scheduled for CY05, with production units to be available in CY07. The company is doing component and subsystem research now to determine the best system trade-offs for a machine design. Next year, Canon expects to begin design of the introductory units. Fabrication, integration, and first shipments are currently scheduled for CY03, CY04, and CY05, respectively. Because of the early stage of Canon's project, technical choices are not yet clear. It would be surprising if Canon is not concerned about technical risks that might delay EUV. These risks have probably caused Canon to place the ML2 project ahead of EUV in its lineup.
New approaches are a good sign
One of the strongest features of market economies is that they permit multiple approaches to common problems. Government and consortia-driven processes tend to drive to monocultures and single solutions. If the plans of the VHSIC program had persisted, the industry would have abandoned optical lithography far too soon. But the market prevailed then. Nikon and Canon's departures from the optics forever roadmap is a healthy sign. Market forces may once again drive this industry to the right solution. Novel e-beam architectures such as PREVAIL and ML2 may be the first NGL technologies to become the now-generation technology. It will be interesting to watch. Any veteran of this business knows it is dangerous, often foolish, to place your bets too soon.
Griff Resor received his BS degree in physics from Yale University and his MBA from Harvard University, and is best known for developing the first commercially successful wafer stepper. He also founded MRS Technology Inc. In 1992, he received a Semi award for these contributions to the industry. Resor is now president of Resor Associates, a company that provides business and technology consulting services to high-tech firms. 629 Massachusetts Ave., Suite 200, Boxborough, MA 01719; ph 978/263-7826, e-mail [email protected].