Lithography at the Crossroads

Can One Solution Fit All or Will Segmentation Take Place?

Bijan Moslehi, PhD, is chief technology officer and senior vice president, semiconductor technology research, for The Noblemen Group, a boutique investment banking, strategic advisory, and business development firm. Moslehi has 20 years' experience working in the semiconductor and semiconductor equipment industries. He can be reached at bmoslehi@noblemengroup.com.

Lithography continues to be the most expensive and important technological challenge facing the semiconductor industry. Patterning has always been at the core of chipmaking. Optical lithography tools and processes, materials, and technologies in various forms and incarnations have played an increasingly prominent role in the wafer fab. Over the past few decades, different implementations of optical lithography have enabled scaling of semiconductors to ever-smaller feature sizes—from 200 µm in the early days to at least 32 nm within this decade. During this time, people have periodically predicted optical's demise. Instead, a constant flow of ideas and major practical innovations by countless engineers and scientists have extended the capabilities of the incumbent optical technologies far beyond the limits of what was once considered possible.

At the dawn of the industry in the 1960s, optical lithography was based on contact printing, which gradually reached its practical limits because of defects and damage resulting from the contact between the photomask and the photoresist on the wafer. Proximity printing, which separated the mask from the resist-coated wafer by a few tens of microns, solved these problems. Projection aligners entered the market in the early 1970s, eliminating the need for putting the mask in close proximity to the wafer. Projection aligners used all-reflective optics with a low numerical aperture (NA<0.2) to project and focus the mask images onto the resist. They were designed to work with the same 1X masks used in proximity printing. To extend the technology, both proximity and projection printing later used deep-UV (DUV) illumination.

However, because of the low-NA optics and the 1X mask quality, in the early 1980s most believed that ~1 µm marked the limit for optical lithography. Work was under way to deploy extreme UV (EUV) and soft x-ray illumination in proximity and projection printing systems, but those efforts were undermined by the timely introduction of the reduction step-and-repeat tools (steppers), which continue to be the industry's workhorses.

By the mid-1990s, large-field 248-nm step-and-scan systems (scanners) with improved resolution were adopted by fabs, with current advanced scanners using 193-nm DUV ArF excimer lasers capable of 90- and 65-nm processing. Over the last decade, DUV lithography (DUVL) has been enhanced to where it can pattern feature sizes below the DUV light source's wavelength. Subwavelength lithography has become a reality because of major advancements in tools (scanners, steppers, tracks, metrology) and materials (reticles, resists) using various optical extension and resolution enhancement techniques, including optical proximity correction (OPC) and phase-shift masks (PSM).

However, this phenomenal achievement has been accompanied by escalating costs. With a price of $15 million to $25 million per system, DUVL accounts for a significant portion of a fab's capital budget. Furthermore, with each new node, increasingly complex optical extension technologies have triggered an exponential rise in mask costs.

Since subwavelength lithography using 193-nm DUV scanners will likely run out of steam at the 45-nm node, the development of a host of potential next-generation lithography (NGL) technologies has accelerated. Originally targeted for the 0.13-µm node, NGL is envisioned to ultimately replace optical lithography and enable continued downward scaling of feature sizes. But major DUVL innovations and progress in immersion lithography (IML) have pushed out the NGL timetable.

Hyper-NA IML scanners—where the space between resist and lens is filled with a liquid (e.g., water) having a refraction index higher than air—should extend the life of optical DUVL, providing a more evolutionary, practical solution for the next few technology nodes. Over the past two years, IML has made tremendous progress and appears to be the technology of choice for 45-nm manufacturing, possibly even for 65 nm. Using fluids with a higher refractive index than water and a potential NA of >1.5, IML could be extended to 32 nm, with efforts under way to push it to 22 nm. Recently, critical levels of commercial 90-nm devices, including SRAMs and processors, have been successfully processed with "preproduction" 193-nm IML scanners. In addition, 65-nm imaging capability has been demonstrated, and decreased levels of immersion-related defects have been reported. Early IML tools are planned for delivery in 2006, with production models to be ready by 2007–2008. However, an expected price tag of $25 million to $30 million per tool, combined with skyrocketing mask costs, will make IML an even more expensive process than 193-nm "dry" DUV scanners.

IML's recent rapid progress has somewhat eased the urgency for implementing NGL. Major next-generation candidates include EUV lithography (EUVL), various incarnations of E-beam lithography (including E-beam projection lithography and direct-write [DW] maskless lithography), and a new entry—nanoimprint lithography (NIL). Although there is no NGL consensus, EUVL is considered the front-runner. Technically, its 13-nm EUV light with multilayer reflective optics will address the resolution and depth-of-focus issues. Over the next five years, EUVL must overcome several major technical issues and challenges, including the lack of a good power source and the need for new resist formulations. Also, the projected price of at least $40 million to $50 million per tool will make EUVL prohibitively expensive.

At the recent SPIE Microlithography Conference in San Jose, speakers expressed serious doubts about the future success of EUVL and predicted its ultimate failure on economic and technical grounds. Of course, the technology's proponents strongly disagreed. Extensive research is under way to resolve the many technical issues of EUV. Intel, the technology's most powerful advocate, is targeting 2009 for EUVL insertion at the 32-nm node, a five-node pushout from the original target of 0.13 µm.

E-beam lithography has a long history as a proven technology. Traditionally used for maskmaking as well as low-volume specialty, ASICs, prototyping, and R&D applications, its low throughput has prevented its adoption for high-volume chip manufacturing. Current efforts are focused on improving throughput and solving other technical issues, such as space charge effects. Because of rising mask costs, maskless lithography, using higher-throughput DW multibeam systems, has recently generated interest.

Nanoimprint problems include low throughput, insufficient overlay capability, and high defect levels. The last issue causes the most concern among chipmakers. In addition, NIL still lacks the required support infrastructure. Certain lessons learned from optical contact printing may help the technology overcome some of its difficulties. Until NIL becomes more production worthy, many in the industry will remain skeptical about its viability as an operationally practical option for nanoscale chip manufacturing. Still, the potential for huge cost advantages makes it an intriguing NGL candidate.

Capability, operational efficiency, and economics must drive the final decision on choosing the right lithography technology. Economics must factor in the overall costs, not just the tool costs. This figure includes the total costs of ownership of tools and materials and must take into account the dual cost impacts of yield and productivity. The economic viability of an option for fabs running a high product mix of advanced low-volume products should be examined, particularly in light of drastically rising mask costs. Some of these fabs are adopting a hybrid (mix-and-match) approach for ASICs and prototyping, using DW maskless E-beam for critical layers and DUVL for noncritical layers. In contrast, fabs processing a low product mix of advanced high-volume products are pressing for IML and EUVL. Those selling higher-margin chips are among the strongest EUVL proponents.

These examples point to a potential future of industry segmentation and multiple lithography solutions, with each segment adopting the most economical approach to best suit its needs and operational requirements. This scenario could change with a major breakthrough over the next five years—such as an affordable high-throughput DW E-beam or an economically and technically viable EUVL—much like how the introduction of steppers revolutionized chipmaking in the 1980s.