High-NA EUV lithography: current status and outlook for the future

While engineers at ASML and Zeiss are addressing the technical issues associated with designing and fabricating high-NA EUV exposure tools and optics, there are several other issues of concern to lithographers at the chip companies who need to create integrated patterning solutions. A summary of some of the key challenges are listed in Table III. ^21,22) These are important considerations, since all elements of a lithographic process must be capable in order for the process to be useable in manufacturing. More detail on these challenges is provided in the remainder of this section.

Resists have long been limiting the capabilities of EUV lithography, ²³⁾ and with demands for increased performance to meet the requirements of scaling, resists are expected to continue to be a constraint for EUV lithography, even with continuing advances in lithographic materials. Early in the development of EUV lithography, it was recognized that resolution was being limited by photoacid diffusion during post-exposure bake. ²⁴⁾ Steps were taken by resist manufacturers to reduce photoacid blur, leading to resists with improved resolution. Later, it became appreciated that the resolution of EUV resists can also be limited by photoelectron and secondary electron blur. ^25,26)

It was also understood that line-edge roughness (LER) limited the applicability of EUV lithography. ²⁷⁾ LER is not something that can be predicted by means of continua models, for it is fundamentally a quantum phenomenon. While LER was recognized initially as relevant to optical lithography, ²⁸⁾ it became an issue of greater concern to EUV lithography due to several factors. First, the small size of features targeted for patterning by EUV lithography meant that levels of LER that were of small consequence for optical lithography became very significant for EUV lithography. Moreover, the weakness of EUV light sources motivated lithographers to attempt to minimize exposure doses, which resulted in a sizable contribution of photon shot noise to LER. ²⁹⁾ Studies of LER later led to recognition that stochastic variations cause defects as well as roughness. ³⁰⁾ (Fig. 8) While LER and defects due to stochastic mechanisms are not issues specific to high-NA EUV lithography, they are highly relevant, since high-NA is expected to be used to pattern devices with many transistors, necessitating a decrease to very low defect levels, and small features will require low LER.

Zoom In Zoom Out Reset image size

As features have continued to scale, the sizes of resist molecules have become non-negligible fractions of critical dimensions. At the molecular level, resist materials are not homogeneous. Inhomogeneity can result from randomness, component segregation and aggregation (Fig. 9). When photoacid generators (PAGs) are not bonded to the resist polymers, PAG-PAG spacings will be variable. For a resist with high PAG loading, the average PAG-PAG spacing is 1.7 nm, ^31,32) which sets the scale for roughness. Variation in component spacings can be increased beyond that expected from pure randomness due to the phenomena of segregation and aggregation. PAGs are polar molecules, so they can attract each other (Fig. 10), resulting in aggregation. With aggregation, the effects of resist inhomogeneity are increased. ³³⁾ Base quenchers are subject to similar variability in distribution.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Segregation in resists has been observed, particularly in ultrathin resist films, even though the mechanisms causing segregation are not yet well understood. Gradients of photoacids from the top to bottom of resists films were inferred from simulations, where observed results could not be explained through models that assumed uniform distributions of photoacids. ³⁴⁾ More recently, segregation in resist films has been measured directly using resonant soft X-ray reflectivity (RSoXR), ³⁵⁾ where data were inconsistent with models that assumed resist material homogeneity. On the other hand, models could be fit well to data if the resist was assumed to consist of three layers of differing composition. (Fig. 11) The problem of segregation becomes increasingly significant as resist films become thinner. At the dimensions of interest for high-NA EUV lithography ultrathin resist films are needed to mitigate pattern collapse. Moreover, as will be discussed in Sect. 3.3, ultrathin resist films will be needed to address small depths-of-focus, making these issues of particular importance for high-NA EUV lithography.

Zoom In Zoom Out Reset image size

In addition to the spacing between components, there are other factors that set the length-scale for roughness in resist patterns. Resist films are made of molecules, which are of finite size. For many years, the size of resist molecules was considerably smaller than minimum device features, so the detailed structure of resists was not relevant to process control. LER became a subject of interest around the same time that ArF lithography was being introduced, although it was then having only a small impact on process control. As features continued to shrink below 100 nm, LER began to affect linewidth control and device performance. ^36–39) Relative length scales for 10 nm ½-pitch technology are shown in Fig. 12. ⁴⁰⁾ As can be seen, process control requirements for high-NA EUV lithography are similar in size to the molecules that comprise resists.

Zoom In Zoom Out Reset image size

Chemically amplified resists based on photoacid-induced deprotection are also prone to pattern collapse at tight pitches. ⁴¹⁾ Considerable effort has been expended to mitigate pattern collapse, but the problem appears to persist at ½-pitches less than 12 nm. Accordingly, pattern collapse is a major issue at the dimensions of interest for high-NA EUV lithography. A detailed understanding of the mechanical properties of resist materials also requires considerations at the molecular level. The results of nanoscopic mechanical calculations of the C₃₃ component of the local elasticity matrix for resist patterns are show in Fig. 13. The outer layers of the features shown in this figure have lower rigidity than the bulk material, resulting from greater free volume for material on the free surface. ^42,43) The fraction of the feature volume composed of this lower rigidity material increases as shapes get smaller, which leads to greater susceptibility to pattern collapse and a need to reduce aspect ratios (height/width).

Zoom In Zoom Out Reset image size

One resist issue that appears to be common to all resist material platforms is blur. The radiation chemistry in EUV resists is driven by photoelectrons that are generated by the absorption of EUV photons, and from resulting secondary electrons, rather than directly from the photoabsorption event. This appears to be a characteristic common to many different types of EUV resist platforms. Understanding the impact of electron blur requires knowledge of the minimum energy at which the secondary electrons can cause a radiation chemistry event and the ranges over which these electrons travel before they scatter and have insufficient energy to induce chemical reactions. This is currently a topic of active research. ⁴⁴⁾ Current estimates for the distances from the initial photoabsorption over which secondary electrons can induce chemical events is ∼3 nm, ⁴⁵⁾ although there are uncertainties to these quantities, which are also dependent on resist material composition. In addition to image blur due to electrons, there is additional blur in chemically amplified resists due to photoacid diffusion during post-exposure bake.

The net result of electron scattering and photoacid diffusion is blur, but these phenomena have statistical characteristics that become significant when there are a small number of events. Consider photoacid diffusion. Individual photoacids will undergo random walks, on-average resulting in blur. (Fig. 14) Electron scattering is similar, though more complicated, because scattering events can result in production of additional secondary electrons. ²⁶⁾ Although the probability is very low, there can be situations where random walks go in one general direction, which can lead to defects. ^6,46) Since defect levels need to be sub-parts-per-billion, very low levels can reduce yield.

Zoom In Zoom Out Reset image size

New types of resist have the potential to overcome many of the issues seen currently with chemically amplified resists. For example, metal-oxide resists, upon exposure lead to networks of cross-linked structures. ⁴⁷⁾ This cross-linking can increase rigidity, thereby reducing pattern collapse. Another type of resist is based upon the scissioning of polymers. The products of resist scissioning are smaller than the original polymers, reducing the scale for roughness at the molecular level. ZEP520A, an electron-beam resist, is an example of a scissioning resist, and there are currently efforts to optimize this resist platform for application to EUV lithography. ⁴⁸⁾

Development of resists targeting 10 nm ½-pitch and below is proceeding. This includes efforts to extend chemically amplified resists as well as the development of alternative platforms. However, resists with low LER, low levels of stochastics-induced defects, no pattern collapse, and the requisite resolution, while avoiding excessively high exposure doses have not yet been found. Improvements in resists will be needed to enable fully high-NA EUV lithography.

Regardless of resist platform or specific formulation, there is a dose below which the photon shot noise induced LER and defects are too large to meet technology requirements. In the 2021 International Roadmap for Devices and Systems, it is projected that doses slightly greater than 80 mJ cm⁻² will be needed to support 10 nm half-pitch patterning. ^49,50) This necessitates light sources with sufficiently high output power to avoid low throughput and productivity. This problem is illustrated in Fig. 15. ⁵¹⁾ If we assume a 500 W source and exposure dose of 80 mJ cm⁻², then the throughput for a 0.55 NA tool will be 120 wafers h⁻¹, substantially less than the 185 wafers h⁻¹ that occurs when the source power is high enough for throughput to be limited by mechanical considerations only. Such a limitation would occur for source powers of approximately 1.3 kW or higher if the exposure dose is 80 mJ cm⁻².

Zoom In Zoom Out Reset image size

ASML San Diego has achieved 400–500 W output with their laser-produced plasma (LPP) light source. ¹⁶⁾ Increases in the output of LPP sources have been achieved through higher efficiency in converting infrared laser light to in-band EUV light, as well as improvements in pulse-to-pulse repeatability. As efficiency and repeatability approach possible maxima, further increases in LPP light output can be achieved through the use of higher power infrared lasers.

Free-electron lasers are being considered as alternatives to LPP light sources. Such sources of light are based on particle-accelerator technology. Electrons that are near the speed of light are injected into an array of magnets (Fig. 16), where the oscillating electrons emit light of wavelength λ, given by the following expression ⁵²⁾:

$$\begin{eqnarray}&&\lambda =\displaystyle \frac{{\lambda }_{u}}{2{\gamma }^{2}}\left(1+\displaystyle \frac{{K}^{2}}{2}\right),\end{eqnarray} \tag{ 2 }$$

where

$$\begin{eqnarray}&&K=\displaystyle \frac{e{B}_{0}{\lambda }_{u}}{2\pi mc},\end{eqnarray} \tag{ 3 }$$

where B₀ is the peak magnetic field in the magnet array, m is the mass of an electron, c is the speed of light, and γ is the relativistic factor:

$$\begin{eqnarray}&&\gamma =\displaystyle \frac{1}{\sqrt{1-{\left(\displaystyle \frac{v}{c}\right)}^{2}}}.\end{eqnarray} \tag{ 4 }$$

A magnet array with a period of 2 cm, with 800 MeV electrons, can produce 13.5 nm EUV light. ⁵³⁾ A cavity with gain can be produced by placing annular mirrors at both ends of the magnet array. ⁵⁴⁾ Free-electron lasers have been made at infrared and other wavelengths, and research on such lasers at EUV wavelengths is being conducted. ^55,56) Although free-electron lasers comprise a substantially different class of light sources than used currently for EUV lithography, they represent a potential long-term path to very high source source power.

Zoom In Zoom Out Reset image size

Focus control has long been a challenge for optical projection lithography. ⁵⁷⁾ At NA = 0.55, the Rayleigh depth-of-focus (DOF) is

$$\begin{eqnarray}&&{\rm{DOF}}=\displaystyle \frac{\lambda }{N{A}^{2}}\end{eqnarray} \tag{ 5 }$$

$$\begin{eqnarray}&&=\,\displaystyle \frac{13.5}{{\left(0.55\right)}^{2}}\end{eqnarray} \tag{ 6 }$$

$$\begin{eqnarray}&&=\,45\,\mathrm{nm},\end{eqnarray} \tag{ 7 }$$

which is about 1/3 that of 0.33 NA tools. With resolution enhancement techniques, depths-of-focus greater than the Rayleigh DOF are possible, but it will remain the case that the DOF decreases substantially as the NA increases from 0.33 to 0.55. Accordingly, much improved focus control will be needed to enable high-NA EUV lithography.

To obtain good imaging, a quality aerial image is needed throughout the depth of the resist. In optical lithography, this is facilitated by refraction that occurs at the air-resist or water-resist interface. (Fig. 17) The relationship between θ₁ and θ₂ is given by Snell's law:

$$\begin{eqnarray}&&\displaystyle \frac{\sin \,{\theta }_{1}}{\sin \,{\theta }_{2}}=\displaystyle \frac{{n}_{{\rm{resist}}}}{{n}_{{\rm{air}}/{\rm{water}}/{\rm{vacuum}}}},\end{eqnarray} \tag{ 8 }$$

where n_resist is the index of refraction of the resist and n_{air/water/vacuum} is the index of refraction of the medium above the resist. Larger values for n_resist lead to greater depth to which acceptable blur can be attained. However, for EUV lithography, the real part of the index of refraction for most materials is close to 1.0 and is often < 1.0. This means that refraction of EUV light is small, and to the extent that there is refraction, it decreases the depth over which the blur is acceptable. As a consequence of the small optical depth-of-focus, coupled with the lack of mitigation by refraction, it will be important to have very thin layers of resist for high-NA EUV lithography. As discussed in the previous section, there are phenomena, such as component segregation, that degrade patterning significantly as resist films become very thin.

Zoom In Zoom Out Reset image size

Focus control is not a problem solely within lithography. Excellent wafer flatness will also be required, starting with bare silicon wafers and continuing through wafer processing. This places tight requirements on film deposition and particularly on chemical-mechanical polishing (CMP). Substrate uniformity will help with focus control, but focus sensors that are very insensitive to substrate films will also be needed.

At high numerical aperture, image contrast is dependent on polarization of the illumination. ⁵⁸⁾ As shown in Fig. 18, image contrast of two interfering plane waves can be maintained with large NA with S-polarized light, but contrast becomes smaller with P-polarized and unpolarized light as the NA is increased. For this reason, the illumination systems of advanced immersion scanners provide polarization control. ⁵⁹⁾ The laser-produced plasma (LPP) light sources planned for initial use on high-NA exposure tools produce unpolarized light. As can be seen in Fig. 18, there can be appreciable loss of image contrast at 0.55 NA with unpolarized light. Polarizing the light involves loss, which would exacerbate all issues associated with low source power. On the other hand, the emission of free-electron lasers is polarized, giving additional motivation to consider free-electron lasers (FELs) as sources for high-NA EUV exposure systems.

Zoom In Zoom Out Reset image size

The reduction of image contrast with unpolarized light is another problem that is mitigated in optical lithography by refraction at the top surface of resist. Since there is no such benefit for EUV lithography, the degradation of contrast for unpolarized illumination occurs at a lower NA than occurred for optical lithography. Fortunately, for the initial applications of high-NA EUV lithography, light rays at the largest angles supported by the numerical aperture will not be essential, so the problem of image contrast loss with unpolarized light will not be a problem at first.

For EUV lithography, computational accuracy requires consideration of many physical phenomena, all of which need to be addressed simultaneously (Table IV). ⁶⁰⁾ Although this complexity precedes high numerical aperture, greater accuracy is required for high-NA because of the smaller feature sizes of interest for high-NA EUV lithography. Moreover, the small depths-of-focus associated with high-NA increase the demands on computational solutions.

In Fig. 19, an example is shown of how computational lithography can be used to improve patterning. ⁶¹⁾ Illumination in one case (Low DOF) has maximum normalized image log-slope (NILS) at best focus, while another illumination has lower NILS at best focus, but constant NILS through focus (Iso NILS). As can be seen in Fig. 19, considerably reduced pattern collapse was seen with the Iso NILS illumination.

Zoom In Zoom Out Reset image size

Inverse lithography technology (ILT) is a technique that has proven useful for constructing mask layouts that maximize process windows. ⁶²⁾ Until recently, because of long computation times, the application of ILT has been limited to addressing hotspots or memory cells. Moreover, optimum mask layouts resulting from the application of ILT have often been curvilinear, which has made it difficult to fabricate the masks. There has been much recent progress to enable ILT fully. GPU-acceleration has reduced computation times significantly, enabling the application of ILT to full-chip layouts with practical computation times. ^63–65) Multibeam mask writers have made it possible to produce high quality masks with curvilinear patterns, ⁶⁶⁾ and an initiative has started to standardize data formats for curvilinear features on masks. ⁶⁷⁾ With capability established for writing masks with curvilinear features with reasonable write times and pattern quality, other elements of mask-making relevant to the fabrication of such masks are being addressed. In terms of EDA infrastructure, software for mask process corrections (MPC) ⁶⁸⁾ and mask rules check (MRC) ^69,70) need to be adapted for application to curvilinear features.

ILT and curvilinear features are not unique to high-NA EUV lithography, but these hcapabilities are coming to maturity in the same time frame that high-NA EUV tools will become available. Consequently, these approaches can be expected to be important elements of high-NA technology. These techniques may also prove useful for extending 0.33 NA EUV lithography, and so could affect the timing for the introduction of high-NA EUV lithography into HVM.

From the perspective of high-NA EUV lithography, there were a number of fundamental mask issues, preceding those discussed in the prior section, that needed to be addressed early in the development of high-NA lithography. Because of ASML's decision for high-NA EUV masks to have the same mechanical form factor (152 mm × 152 mm × 6.35 mm) as those used currently for 0.33 NA EUV and advanced optical lithography, the tools for making high-NA EUV masks can be the same ones as used currently, at least with modest improvements. This greatly simplifies mask making for high-NA EUV lithography. Given the length of time needed to create a complete mask-making infrastructure for masks with a different format (such as 230 mm), ⁷¹⁾ it is likely that the development and eventual insertion of high-NA EUV lithography would be gated at many steps by mask availability. Although there are some disadvantages to using 152.mm × 152 mm × 6.35 mm substrates for high-NA EUV lithography, particularly those discussed in the next section, the decision to stay with this form factor was a practical one.

Tantalum-based absorbers for EUV masks enabled the development of EUV lithography and its initial use in HVM. ^72–74) However, with dimensions shrinking, mask 3D effects are becoming increasingly significant when such absorbers are used. ⁷⁵⁾ Alternative mask architectures have been considered, ⁷⁶⁾ but most recent effort has gone into identifying new mask absorbers for both binary and attenuated phase-shifting masks (attPSM). ⁷⁷⁾ For binary masks, mask 3D effects can be reduced by having the real part of the index of refraction of the absorber be close to 1.0, while the extinction coefficient (k = imaginary part of the index of refraction) should be as large as possible. This latter condition enables the use of thin absorbers.

For attenuated phase-shifting masks, the desired phase can be achieved with thinner absorbers when the real part of the absorber's refractive index differs significantly from 1.0. ⁶⁾ Since it is desirable to have an appreciable amount of reflection from the absorber on an attPSM, the absorber should have a moderate extinction coefficient. This means that candidate elements for the absorbers of attPSMs should be those in the lower left corner of Fig. 20, although they may be alloyed with other elements. Since rhodium is an extremely rare element, all isotopes of technetium are radioactive, and there is a dearth of volatile compounds of palladium, absorbers for attenuated phase-shifting masks need to contain substantial amounts of ruthenium. This has implications for the capping layer, which has long been composed of ruthenium, since the capping layer has also provided etch selectivity to the absorber. Accordingly, use of absorbers containing a significant amount of ruthenium will also require an alternative capping layer.

Zoom In Zoom Out Reset image size

Because mask absorbers need to meet a large number of requirements (Table V), identification of suitable materials is challenging. ⁷⁸⁾ Ruthenium is a material known to meet all requirements, and attPSM development has been progressing well. ⁷⁹⁾ Identification of a suitable material for a high-k absorber has proven to be more challenging, but development is proceeding. ⁷⁸⁾

Attenuated phase-shifting masks are being developed long after the first integrated devices were fabricated using EUV lithography. ⁸⁰⁾ This delay was due in large measure to the difficulty in finding theoretical solutions. For example, it was only recently understood that the ideal phase shift for EUV attPSMs is not 180° = π as in optical lithography, but is closer to 1.2π. Moreover, optimum solutions may be pitch-dependent. Identifying theoretical solutions is being pursued in parallel with the development of attPSM mask blanks. ⁸¹⁾

Alternative mask absorbers are discussed in the context of high-NA EUV lithography, primarily because they will become available at roughly the same time as high-NA tools, and not because they are necessarily integral or specific to high-NA. Roughness is another aspect of masks that is relevant for 0.33 NA EUV lithography as well but is much more significant for high-NA. The extent to which absorber edge roughness transfers from mask to wafer is dependent on lens resolution, as higher spatial frequency roughness will transfer at larger NA. ⁸²⁾ (Fig. 21) Similarly, mask multilayer roughness, which has been shown to have a measurable, but not large impact on local dose variation for 0.33 NA optics, ⁸³⁾ appears to be more significant for high-NA EUV lithography. ⁸⁴⁾

Zoom In Zoom Out Reset image size

The use of anamorphic lenses necessitates commensurate dimensional changes on the masks. The 8× reduction relaxes dimensions along one axis, but dimensions along the other axis will be sized 4×. Since patterning, inspection, and measurement requirements are driven by the smallest features, much mask making technology will continue to be driven by a continuation of scaling requirements. For inspecting patterned high-NA EUV masks, Lasertec has been developing an extension of their ACTIS A150 actinic patterned mask inspection tool, ⁸⁵⁾ with higher resolution optics and different resolution scales in the x- and y-directions. Additionally, die-to-database algorithms, suitable for the asymmetric configuration of high-NA, are being developed. These algorithms will also need to be ultimately capable of handling curvilinear features.

Integrated circuits with very large die sizes have become important products. Integrated circuits such as the Nvidia GA100 GPU and Intel Skylake processors are so large that only one die can fit in a 26 mm × 33 mm exposure field, and such devices are too large to fit in the ½-field of high-NA exposure tools. In order to continue producing devices with similar chip sizes, stitching will need to be adopted. With stitching, part of the chip layout is patterned using one mask, while the remainder is patterned by exposure of a second mask (Fig. 22).

Zoom In Zoom Out Reset image size

Stitching is not a new lithographic technique, having long been an element of electron-beam lithography, and it has also been used in optical lithography for the production of large-area devices. ⁸⁶⁾ However, very precise implementation is necessitated by the tight process control requirements of the nodes at which high-NA EUV lithography will be applied. There are also special considerations that apply to EUV lithography. For example, flare (scattered light) is significantly higher in EUV lenses than optical lenses. As a consequence, scattered light from the exposure of one sub-field can expose resist in the area allocated to the other sub-field, with a range of microns. Another set of issues arises from the black border on EUV masks. Because EUV absorbers are not fully effective in suppressing reflected light, the multilayer reflector around the exposure area on the reticle is often removed by etching (Fig. 23). Mo/Si multilayers have 100–500 GPa stress. ⁸⁷⁾ Etching the black border leads to locally reduced stress, which can displace mask features. ⁸⁸⁾ The results of careful measurements performed at Imec of pattern displacement are shown in Fig. 24. ⁸⁹⁾

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In addition to pattern placement errors, there can be aerial image cross-talk across the stitch boundary, as well as variations in flare due to different pattern densities on the two sides of the boundary. These phenomena that affect critical dimensions can be compensated through OPC, and pattern placement errors can also be corrected by calculating the displacement due to stress reduction and compensating during mask writing. Because the magnitude of these effects is greatest near the stitch boundary and the accuracy of such corrections may become inadequate very close to the boundary, a 1 μm exclusion zone between sub-fields has been proposed. ⁸⁹⁾ While such an exclusion zone will have little impact on the number of dies/wafer, it does preclude geometries that cross stitch boundaries, such as interconnects. Fortunately, the tightest pitch metal layers are usually confined to logic and memories cells and local routing. These can be done with high-NA EUV lithography, while the routing that needs to cross stitch boundaries can be patterned using 0.33 NA EUV lithography. It is also reasonably straightforward to avoid having contacts and vias within this exclusion zone. The special issues associated with stitching provide additional motivation for using multiple patterning with 0.33 NA exposure tools rather than high-NA EUV lithography, particularly for across-die routing.

An alternative solution is the avoidance of large dies altogether. With advanced packaging, it has proven possible to maintain high performance even while separately fabricating parts of circuits that previously had been fully integrated into a single die. ^90,91) This is currently being done to improve the cost-effectiveness of advanced lithography. ⁹²⁾ For example, the high performance logic circuits of AMD's Epyc processor have been fabricated using 7 nm technology, while input/output functions were fabricated using lower cost 14 nm processes. ⁹³⁾ In other applications, memory access rates are improved using advanced packaging technology. ⁹⁴⁾ It may also be possible to achieve high levels of performance with maximum die sizes that can fit into the exposure fields of high-NA EUV exposure tools.

As described in Sect. 2, high-NA EUV exposure tools will be complex, and fabrication will require exceptional attention to detail. As a consequence, high-NA EUV exposure tools are expected to cost in excess of $300 M. ⁹⁵⁾ Over many prior generations of exposure tools, the exponential increase in prices was offset by improvements in throughput, ⁹⁶⁾ and with high capital costs for high-NA EUV tools, throughput will be especially important.

The projected throughput of first generation high-NA EUV exposure tools, as a function of source power divided by exposure dose, was shown in Fig. 15. For low source power or high exposure doses, the throughput is less than the maximum, which is limited by the mechanical capabilities of the system. Source power of 400–500 W has been achieved by ASML San Diego for LPP sources, with apparent plans for even higher power. In the 2021 IRDS, for the "2.1 nm" node, which will involve a 20 nm minimum metal pitch and be in production in 2025, exposure doses of 80 mJ cm⁻² are predicted. With >800 W sources, the throughput of high-NA EUV exposure tools will be close to the mechanical limit.

The throughputs of Fig. 15 are based on an assumption of 96 shots/wafer for 0.33 NA tools and 188 shots/wafer for 0.55 NA systems. In wafer fabs, the number of shots per product wafer is usually greater than the number of shots on which exposure tool suppliers' throughput specifications are based, so actual throughputs in production will be less than those indicated by the graph of Fig. 15. In one study, actual die sizes were used to determine the number of shots per wafer for high-NA EUV tools, resulting in the distribution shown in Fig. 25. ⁹⁷⁾ In all cases the number of shots per wafer exceed 188, and for several die sizes, the number of shots per wafer was more than twice that amount. With more shots per wafer than assumed for the throughputs shown in Fig. 15, scanner throughput in production will generally be lower than those in the graph.

Zoom In Zoom Out Reset image size

Besides requiring more shots per wafer, there additional ways in which short exposure fields increase the difficulty of attaining high throughputs. In scanning, the reticle and wafer stages are accelerated until they reach the proper scanning speed, which needs to occur before the top part of the slit begins to expose the wafer, and it continues until the bottom part of the slit no longer is exposing the wafer. (Fig. 26) Thus, the total scanning distance is longer than the field height H_F by an amount Δ. This constraint is true on optical scanners, but the extra scanning distance Δ on EUV tools is made longer by the curvature of the slit. The scan time t_exp is given by

$$\begin{eqnarray}&&{{t}}_{\exp }=\left(\displaystyle \frac{S}{\bar{I}H}\right)\left({H}_{F}+{\rm{\Delta }}\right),\end{eqnarray} \tag{ 9 }$$

where S is the exposure dose and $\bar{I}$ is the average light intensity across the slit. If we assume that high-NA tools will have the same slit height H and curvature as 0.33 NA tools, then scanning the distance Δ takes approximately ¼ of the total scan time, double what is necessary on 0.33 NA tools. With half-height exposure fields, it will be challenging to attain cost-effective throughput for high-NA EUV exposure systems.

Zoom In Zoom Out Reset image size

Reliability and maintenance also affect the cost-effectiveness of lithography. Equipment downtime directly reduces tool output, while maintenance costs include the cost of replacement parts and wages for service technicians. Reliability and maintenance costs of high-NA exposure systems are being mitigated by ASML's decision to incorporate certain modules, such as wafer and mask handlers, that are used in their 0.33 NA tools. The use of identical modules will take advantage of past reliability improvements and reduce costs for stocking replacement components. Many years of experience with laser-produced plasma light sources can be applied to high-NA, accelerating the time to achieve high uptime.

Cost has always been a concern of lithographers, and it will certainly be a challenge for high-NA EUV lithography. Higher power light sources will be beneficial. In addition to providing polarized light, free electron lasers are expected to have high power, so cost is another reason for free electron lasers to be given consideration as sources for EUV exposure tools.