Extending 193 nm Optical Lithography

Optical lithography continues to be the engine that powers Moore's Law. Recent extensions of 193 nm (ArF) lithography using resolution enhancement technology (RET) and hyper-NA immersion optics have demonstrated resolution capable of patterning 40 nm half-pitch. This will enable the production of higher-density DRAM and flash memories, as well as new generations of multicore processors. These improvements have come about because of the development of water immersion optics with a numerical aperture (NA) exceeding 1, the use of polarized illumination, and the maturation of several low k₁ technologies, such as model-based optical proximity correction (OPC), source-mask optimization, and various forms of double exposure lithography.

OPC and custom illumination

Image contrast degrades as k₁ is lowered, and consequently the size and shape of the final image also degrades. Short of going to the next NA, the equivalent of increasing the k₁ value, there are two main choices to improve the image: change the shape and size of features on the mask, and optimize the illumination. The first choice is known as OPC, in which the edges of mask features are purposely moved with respect to the CAD layout to achieve an improved final image on the wafer.

Illumination and source-mask optimization

In fabricating memory, however, density is the key and this puts significant restrictions on what options one has for OPC. In this case, illumination optimization is very attractive. Figure 1 shows how the lithography process window for a typical pattern, a DRAM isolation layer, also known as the “brick wall,” can be improved by using an optimized illuminator customized for that layer, compared with more standard illumination types, such as annular. At a further level of sophistication, it is possible to vary mask layout and optimize illumination simultaneously, called source-mask optimization (SMO). In this way, the maximum process window possible can be achieved (Fig. 2 ).¹

1. The lithography process window for a typical pattern, a DRAM isolation layer, can be improved by using an optimized lluminator customized for that layer. In this example, NA=0.8, λ=248 nm and CD=120 nm.

2. It is possible to optimize both mask layout and illumination simultaneously ­— called source- mask optimization (SMO) — to maximize image contrast.

Scattering bars

On the other hand, in logic devices and many practical cases, the device is not laid out on the densest pitch possible and the critical features are often relatively isolated, in an optical sense, from neighboring features. The process window for printing small isolated features is usually worse than that for dense features because of a small depth of focus. A very successful approach to improving the process window for logic structures, such as might be found at the gate level of a system-on-a-chip (SoC) device, involves the addition of scattering bars to the mask.² Scattering bars are sub-resolution mask features that improve the printing fidelity of isolated features by making them appear more dense (Fig. 3 ).

3. A successful approach to improving the process window for logic structures involves the addition of scattering bars to the mask.

Double exposure

Combined with off-axis illumination, scattering bars can increase the size of a lithography process window substantially. Automated software is available to place and size scattering bars on a given CAD layout, as well as recommend optimized NA and illumination parameters for scanner settings. Currently, the industry is researching another method of reducing k₁ in production: double exposure (DE). An example of DE, double dipole lithography (DDL), is given in Figure 4 .³ The idea here is to take a given mask pattern and break it down into two layers — one with critical features all aligned in the X axis, and one with critical features aligned in the Y axis. The first mask can be effectively exposed using an X dipole diffractive optical element, the second with a Y dipole. The combined image can achieve better resolution and image fidelity than would be possible in a single exposure on the same scanner. This is an exciting new area to explore in k₁ reduction, and is the source of current research at ASML.

4. Double dipole lithography (DDL) breaks a mask into two layers, one with critical features aligned in the X axis, and one with critical features aligned in the Y axis.

Hyper-NA immersion lithography

The other exciting new direction in which 193 nm lithography has been extended is through the use of water immersion optics.⁴ As noted earlier, it is possible with water to design optical systems with NA>1. ASML first achieved this with its Twinscan XT:1700i,⁵ which has a maximum NA of 1.2, thanks to its catadioptric lens. Development of hyper-NA optics have stressed the size and weight of optics. To date, most scanner optics from ASML and its competitors have been dioptric designs, in which only refractive elements of either fused silica or CaF₂ have been used.

To control the complexity and cost of the projection optics, ASML's optics partner Carl Zeiss has designed the 1.2 NA lens to be catadioptric, using both mirrors and refractive elements. This has allowed the large aperture to be built within the same form factor as that of previous lenses, thus requiring no major changes to the Twinscan platform.

5. Carl Zeiss’s 1.2 NA lens is an inline catadioptric design that offers 15% more NA while using 40% less glass than a dioptric design.

To modify the Twinscan for immersion, some aspects of this architecture remain intact while others needed to be changed. This summer, ASML announced its fifth-generation immersion tool, the XT:1900i, which features consistency in both lens design and tool-body construction. ASML was not forced into a major lens redesign to achieve an NA of 1.35, or 40 nm resolution.

It is a feat of modern fluid dynamics and mechanical engineering to enable high-speed scanning of wafers through water, scanning with a velocity >500 mm/sec, while maintaining both diffraction-limited imaging and fine overlay. We have also been able to meet overlay requirements of <8 nm.

Exposing resists in water also brought new challenges to scanner design. Scanner suppliers had to develop high-productivity systems for both hydrophobic and hydrophilic surfaces. Immersion lithography must also achieve the same low level of defect density as dry lithography. Over the past few years, concerns for potential new sources of defects such as bubbles of air in the water, resist effluents, water drying stains, and so forth have been studied in great detail.

Substantial progress has been made, however, in both understanding new defect mechanisms and reducing or eliminating these sources. In the end, the most important evidence that immersion is working at acceptable defect levels has come from announcements that highly complex ICs have been successfully fabricated using immersion lithography,^6,7 and its implementation in the production of advanced chip designs this year.

The likely future

Over the next few years, 193 nm lithography, buttressed by the twin pillars of hyper-NA immersion optics and low-k₁ technology, will continue the reduction shrink required by leading volume producers of ICs. But just how far will it go?

The refractive index (RI) of water establishes the upper limit of NA~1.4 for water immersion lithography. But an aperture of at least 1.6 is required to produce 32 nm design rule devices with a k₁ of ~0.28, the current benchmark for mass production with good yields. Consequently, apertures as high as 1.6 will require the introduction of both new fluids and new glasses with a higher RI.⁸

Progress has been made by research programs in identifying several fluids with an RI~1.65. However, an immersion fluid also has to satisfy a number of other physical and economic requirements as well. For example, it must have low absorption, low index variation with regard to temperature (dn/dT), and viscosity and wetting characteristics suitable for high-speed scanning, and it must be readily available for relatively low cost. When one considers this set of stringent requirements that must be met, one is struck by what a miracle fluid water is.

Likewise, research programs are underway to find new glasses for the bottom lens element, which could replace fused silica, which has an RI of 1.57, or CaF₂, which has an RI of 1.50. Current focus is on materials such as ceramic spinel, with an RI of 1.9, and garnets (e.g., LuAG) with an RI of 2.1. But LuAG also has high intrinsic birefringence (IBR), a material property that makes light polarized along one axis travel slower through the material than light polarized along an orthogonal axis. Such high IBR leads to unacceptable imaging degradation, requiring greater lens design complexity to compensate for these effects. Likewise, there are issues with the ceramic material; its microstructure leads to a relatively high amount of stray light. Optimization of grain sizes and structures would be required to minimize the consequent contrast loss. For both candidates, new or improved processes are also needed to fabricate lithography-grade materials.

The business model for new fluids and glasses is also of concern. The worldwide demand for a specialized garnet material used exclusively for the last lens element in hyper-NA 193 nm scanners might only be a few hundred kilograms stretched over several years, but the investment required to produce such materials in the required optical grade would nevertheless be very substantial. It is not clear if any glass supplier has the ability or constitution to take such a risk.

What are some other ideas that may allow further extension of 193 nm lithography? One idea that has attracted attention in the past couple of years is the use of DE and double patterning technologies. As noted previously, the use of two exposures can enable a smaller k₁ to be achieved, such as with DDL. However, it is not possible to resolve features at a k₁<0.25, even if two exposures are used in a single resist step, since the combined image has zero contrast. However, in double patterning, one can effectively produce features with a k₁<0.25. An example of this process and experimental results indicating resolution of grating patterns down to k₁=0.19 has been demonstrated at ASML (Fig. 6 ) and IMEC.⁹

6. Double patterning effectively enables the printing of features with a k1<0.25. (Source: ASML, IMEC)

The advantage of this double patterning technology is that it is a relatively straightforward extension of existing technology — no new fluids, glasses, reticles, exposure sources, resists, etc. are needed to make it work. However, there are several disadvantages: Two (or more) reticles are needed for each critical layer, and twice the number of exposures must be done, effectively cutting scanner productivity by half. Thus, it is clear that, regardless of what path it takes, the cost of ownership of advanced lithography will be challenged.

In addition, double patterning brings with it very tight overlay requirements, since overlay errors are now directly part of the critical dimension (CD) budget. If, for example, half the usual CD budget of 10% of the minimum feature is given to overlay, this implies an overlay budget of R/20. If we challenge ourselves to extend 193 nm to the 32 nm design rule, this implies an overlay budget of <2 nm, roughly 5× tighter than is possible today. A further critical issue is the availability of sophisticated design-splitting software, which can deconstruct a complicated IC into two complementary patterns that can be recombined through double imaging without errors.

So, we face considerable challenges for the further extension of 193 nm lithography beyond 40 nm half-pitch memory designs in both the development of high-index materials to drive apertures beyond 1.35, and in further extension of low-k₁ technology through DE/double patterning. Customer requirements and the momentum of recent developments makes it likely these further extensions will be attempted, however. Indeed, the point at which we have to give up and say we need to move to the next wavelength, 13.5 nm or extreme ultraviolet (EUV), will be the next great debate in lithography.

Acknowledgements

The author would like to acknowledge the many engineers at ASML and Carl Zeiss who work on the extension of 193 nm technology through low-k₁ enhancements and the development of hyper-NA immersion systems.