Special Section on EUV Sources for Lithography

Development of laser-produced plasma sources for extreme ultraviolet lithography

[+] Author Affiliations
Gerry O’Sullivan, Bowen Li

University College Dublin, School of Physics, Belfield, Dublin 4, Ireland

J. Micro/Nanolith. MEMS MOEMS. 11(2), 021108 (May 29, 2012). doi:10.1117/1.JMM.11.2.021108
History: Received July 28, 2011; Revised October 12, 2011; Accepted January 3, 2012
Text Size: A A A

Open Access Open Access

Abstract.  The current status of laser-produced plasma source development for extreme ultraviolet lithography is reviewed. The advantages of using Sn as a fuel for 13.5 nm are discussed as is the rationale for using Nd:YAG prepulses followed by CO2 irradiation of mass-limited Sn droplet targets. To date CEs of around 6% have been obtained in the laboratory while much lower values closer to 2% have been achieved in high repetition rate industrial sources suitable for use in high-volume manufacturing (HVM). The discrepancy appears largely to arise from the mismatch between the effective target lifetime and pulse duration as well as incomplete vaporization or fragmentation of the droplets. Improvements in laser technology and droplet delivery systems should see a marked improvement in the near future, so that CEs of 5% to 6% should eventually be realized in industrial sources. To keep pace with Moore’s law, development work has begun on sources operating at 6.x nm, the wavelength selected for lithography beyond 13.5 nm. It is shown that Gd plasmas with an electron temperature close to 110 eV will provide the solution at this wavelength and the strongest lines ocurring in a Gd plasma are identified.

Figures in this Article

The earliest application of extreme ultraviolet (EUV) emission from laser-produced plasmas was as sources for absorption spectroscopy after the spectra of plasmas of elements with 62Z74 were found to contain broad regions of line-free continua in the EUV and soft x-ray range.1 It was noted that in the lower Z elements of the sequence, an intense modulation of the continuum appeared in the 5 to 6 nm region that moved to shorter wavelength with increasing Z. A similar feature was noted in the spectra of all elements from tin through the lanthanides. Moreover, it was seen that if the concentration of the high Z element was reduced in the target, the feature actually grew in intensity and became significantly narrower while the background continuum intensity significantly decreased.2 The plasmas were produced by a Q-switched ruby laser and the power density on target was 1011W cm2, giving an average plasma temperature of 55eV. Shortly afterward, the application of these plasmas as irradiance transfer standards in the EUV was established.3,4 In the course of this work comparison was made with the emission from CO2 laser-produced plasmas and the results showed that line emission in the latter was noticeably stronger due to the lower plasma density and thus lower opacity and also that the intensity could be enhanced by using a double pulse or pre-pulse illumination. Subsequently, the application of laser produced plasmas as sources for lithography was first demonstrated using emission from a Nd:YAG irradiated steel target.5

Under the experimental conditions of this early work, the maximum ion stage was estimated to be 15 or 16 times ionized and in the absence of self-absorption, calculations assuming collisional radiative (CR) equilibrium6 predicted that the spectra should be dominated by line emission. However, the variation of spectral profile with concentration, along with the observation of strong absorption lines, indicated that self-absorption needed to be taken into account and the reduction in continuum intensity could be traced to the lowering of the contribution from recombination radiation, due to the lower average charge in plasma where the major constituent was a low Z material since the intensity due to recombination scales as z4 where z is the ionic charge. Thus opacity was a limiting factor whose impact can be reduced by reducing the plasma density either through reducing the concentration of the element of interest in the target, using CO2 rather than solid state lasers or using prepulses and so interacting with lower density target plasmas. It was shown that the feature arose from 4p64dn4p54dn+1+4dn14f transitions, which merge to form a UTA (unresolved transition array) consisting of tens of thousands of lines, while the peak position is sensitive to atomic number and moves to shorter wavelengths as Z increases; in Sn it lies near 13.5 nm.7,8 Configuration interaction in the final state is very important and leads to a spectral narrowing and, in the heavier lanthanides, better overlap between the emission from adjacent ion stages than would be obtained by consideration of 4d4f or 4p4d transitions alone. As a result, the UTA can be the brightest emission feature in the spectra of plasmas of these elements. The effects of CI are to concentrate the emission intensity in the high energy end of the array and cause the oscillator strength envelope of the entire array to narrow. The strength of this interaction depends on the magnitude of the 4d, 4f overlap integral, 4d|4f, and this factor also controls the degree of overlap in spectra from adjacent ion stages.

For lithography at 13.5 nm, because of the short wavelength, a suitable source is expected to have a relatively high plasma temperature, and so will utilize either discharge or laser produced plasmas. The parameters to be optimized are consequently the plasma composition in terms of elements to be used, electron and ion densities, electron temperature, and duration of the emission. Three candidate elements quickly emerged: Li, Sn, and Xe, all of which have ions with resonance transitions at 13.5 nm. However, the conversion efficiency of Xe-discharge-produced plasma (DPP) sources was found to be at best 0.5% while laser-produced plasmas (LPP) generated on solid Xe targets had a maximum conversion efficiency of close to 1.2% when irradiated with Nd:YAG laser pulses at λ=1.06μm.9 It was shown10 that the 13.5 nm emission from Xe originated from 4d84d75p resonance transitions in Xe10+. The population of this ion stage needs to be optimized to attain maximum intensity. However, the Xe spectrum is dominated by a 4p64dn4p54dn+1+4dn14f UTA near 11 nm, and a subsequent study performed at the NIST electron beam ion trap (EBIT) showed a five-fold enhancement in a 2% bandwidth near this wavelength where emission11 arises from Xe11+Xe17+.

The analogous transitions in Sn lie near 13.5 nm. In 2005, industry set a power requirement of 120 W in-band at intermediate focus as a target for source development. It became immediately obvious that while Xe might have a future as a fuel in metrology sources, for high-volume manufacturing, where EUV power and a high CE were essential, the use of Sn was more suitable. However, radiation transport, in addition to debris mitigation, is a very significant issue. In a controlled study with tin-doped glass targets, summarized in Fig. 1, it was found that the brightness attainable with slab targets at 5% concentration was some 45% greater than with a pure tin target, and yielded12 a conversion efficiency close to 3% with a Nd:YAG pulse focused to a power density, Φ, of 2×1011Wcm2. The transitions responsible for the emission were identified by Churilov and co-workers13 and shown to originate from Sn8+ to Sn12+. Plasma modeling calculations within the collisional-radiative (CR) regime (where collisional ionization is balanced by radiative and three-body recombination) performed to obtain the ion distributions as a function of electron temperature showed that the condition for production of these ion stages for an optically thin plasma corresponds to electron temperatures in the 30 to 50 eV range.14

Grahic Jump LocationF1 :

Comparison of Sn emission from solid and low density targets. The in-band intensity is clearly greatest for the plasma containing a Sn concentration of 5% by number due to reduced opacity.

To minimize debris, mass-limited droplets were proposed and developed as the optimum target configuration for EUV sources while debris mitigation schemes such as plasma curtains to ionize neutral debris and electrostatic and magnetic deflection setups were developed.15,16 Using droplet targets with Nd:YAG irradiation, a number of researchers found that the maximum CE obtainable was also of the order of 2% to 2.5% but was better for shorter laser pulses. The reason that the 6% CE predicted by considering the earlier Xe work was not attained was due to two factors: plasma opacity and suboptimal emitting volume. If targets containing a few % Sn by number were used, the CE increased, though such slab targets are not practical for high-volume manufacturing both because of delivery and thermal issues, while the use of prepulses with 100% Sn containing targets could effectively have the same impact and increase the CE by a factor of up to 80%.17 With a prepulse the plasma has time to expand and the resulting increase is due to the interaction of a laser with a larger plasma EUV-emitting volume at a lower density than produced by the laser solid interaction directly thereby reducing opacity effects. Another method of reducing Sn concentration was the adoption of foam targets where concentrations of 0.5% gave CEs of around 1.5%.18 Various plasma models all pointed to a CE of 5% to 6% being attainable in the absence of absorption, but this figure is reduced to 2% to 3% when opacity effects are included.

In a detailed study using laser irradiation of polymers coated with a 100-nm-thick layer of Sn, Ando et al.19 measured both the angle of the EUV emission and the CE for a number of pulse lengths varying from 1.2 to 8.5 ns. They found that optical depth increased linearly with pulse duration and obtained a maximum CE of 2.2% at a pulse length of 2.3 ns for Φ=5×1010Wcm2. The results also showed that the maximum EUV emission was obtained in the direction of the target normal and that the optimum pulse duration was determined not only by the optical depth but also by the fraction of the laser energy absorbed in the EUV-emitting region, which is located in the lower density corona. Indeed, an earlier theoretical study found that most of the in-band emission originated from the plasma periphery, in this case the outermost four of the 500 cells used to model a 1-D expansion.20 The result of this simulation is shown in Fig. 2. In their paper, Ando et al. calculated the fraction of the laser energy to be absorbed in the EUV emitting region to vary from 25% at 1.2 ns to 85% after 8.5 ns corresponding to absorption at increasing scale lengths due to plasma expansion. Hence the emission can be controlled by increasing the scale length while simultaneously minimizing the optical depth. The use of prepulses provides one efficient method of accomplishing this outcome. An alternative method is to use suspensions or “mist” targets. Aota and Tomie attained a CE better than 4% using a suspension of SnO2 nanoparticles suspended in a 300 μm volume,21 while more recently the advantages of using such distributed or “mist” targets has been shown in terms of better matching of laser pulses to target lifetime and emitting volume.22 For solid targets, it was also shown that the emission was not isotropic but varied with viewing angle and was maximized when the plasma was viewed at normal incidence. Subsequently this effect was studied in more detail23,24 and the results showed that the intensity varied slightly when viewed from 45 deg to normal incidence but dropped to about 50% of this value when viewed parallel to the target surface. So for any illumination setup the plasma should be viewed as close to normal incidence as possible.

Grahic Jump LocationF2 :

(a) Time dependence of the net power output for a peak power density of 31011Wcm2. The dashed curve indicates the laser pulse. In (b) is shown the contribution of the four outermost fluid cells, with the dashed curve indicating the very outermost cell.

Since the critical electron density, i.e., the density at which laser light is most efficiently absorbed depends is approximately 1021λ2cm3, the use of CO2 lasers operating at λ=10.6μm results in plasma electron and ion densities of a factor of 100 lower than for Nd:YAG produced plasmas.25 Calculations by Nishihara and co-workers26 showed that for CO2 radiation incident on a solid Sn target at Φ=1010Wcm2, a conversion efficiency of 6% should be attainable in a 30 eV plasma and that at lower densities and irradiances, even higher CEs may be realized. However, the number of ions becomes so small that the absolute emitted intensity is insufficient even at 100 kHz repetition rates to reach the power requirements for HVM. The calculations also showed that the spectral efficiency reaches a maximum of 35% for at an ion density of (45)×1017cm3 and electron temperature of 45 eV. Under these conditions, the optimum pulse duration is of the order of 10 ns. Early work with CO2 lasers and solid Sn slab targets indicated that conversion efficiencies approaching 4% could be obtained for 10 ns pulses incident on slab targets and that the CE dropped to less than 3% for longer pulses. However, using 100-ns-long CO2 pulses, the CE increased from 2% to 4% if plasmas were formed in 200-μm-deep grooves27 at Φ=(13)×1010Wcm2. Subsequently it was found that the CE increased from 2% to close to 5% after twenty laser shots at the same target position under the same conditions as above,28 while a similar study29 using 30-ns-long CO2 pulses at Φ=6×1010Wcm2 found that the CE increased from 2.7% to 5% after around sixty pulses on the same position. Moreover, the highest CEs were obtained when the groove width was between one and two times the focal spot diameter. Studies showed that compared with a planar target, where the plasma expands freely and the density decreases rapidly away from the surface, a grooved target controls the hydrodynamic expansion through plasma confinement, which leads to a gentler density gradient. The gentler slope results in a more gradual variation of refractive index and less reflection from the plasma. In addition, the geometrical confinement reduces energy loss from the EUV production zone through lateral expansion, thereby enhancing radiation losses.

However, slab targets cannot be used for high-volume production due to the heating produced at the high laser rep rates required of up to 100 kHz. The most widely used solution to date is provided by the use of mass-limited droplets. However, direct interaction of a CO2 pulse with a droplet source will not result in an optimized CE as the plasma density gradient is too steep, the emission is essentially into 2π unlike in a solid target where half of the radiation is reabsorbed and essentially reused. If the droplet is very small, a significant part of the beam may bypass it completely during the initiation phase, as CO2 pulses are typically focused to spot sizes of 100 μm or more.15 To overcome this problem it is essential to first vaporize the droplet using prepulses and ideally the prepulse should be provided by a short wavelength laser that can be focused more tightly and produce a high-density plasma that can expand to reach the optimum density for CO2 interaction when its size matches the diameter of the main pulse. The first successful demonstration26 involved the use of 10-ns Nd:YAG laser pre-pulses incident on 40-m-diameter Sn droplets at Φ=5×108Wcm2. If the main 10-ns CO2 laser pulse arrived 180 ns after the end of the prepulse, a CE of 6.5% was attained in these experiments. Later, using 36 μm droplets, irradiated with Gaussian 1.06 μm pulses at Φ=5×1010Wcm2 to 4×1011Wcm2 and a main 10.6 μm pulse with a duration of 30 to 50 ns and Φ=1010Wcm2, a CE of 4% was measured at an interpulse delay close to 1 μs, at which time the droplet has expanded to fully fill the focal spot of the CO2 laser. At the present time, there are two main contenders for EUV sources at 13.5 nm: discharge-produced plasma (DPP) and LPP. Traditionally, DPP sources are combined with a grazing incidence collector while LPPs are used with normal incidence optics with a collection efficiency of 30%, three times that of the grazing incidence configurations used with DPP sources. So the collector geometry favors the adoption of LPP, and though debris is still an issue, one novel solution that is being tested is the use of liquid metal coated optics using a tin alloy as the liquid metal.30

Currently Gigaphoton has reported the operation of LPP sources producing 42 W (21 W clean) at intermediate focus with 30 μm droplets corresponding to a CE of 2.1% and has also reported a CE of 3.3% using 20 μm droplets and full target vaporization.31 Part of the reason for the lower than expected CE is the longer than optimum duration of the 10.6 μm pulse and incomplete vaporization/fragmentation of the droplets. Novel schemes to extend the plasma lifetime while maintaining the plasma density and correct gradient profile are being investigated in our laboratory and may also be provided by the distributed targets mentioned earlier.21,22 In their first-generation commercial source, Gigaphoton hopes to realize 140 W at IF assuming a 4% CE, while their second-generation source is expected to deliver 250 W of in-band EUV at IF, again with a CO2 irradiated Sn droplet at a CE of 5%.32

In 2009 ASML announced that sources would be needed at 6.x nm for future lithography beyond 13.5 nm predicated on the availability of mirrors, this time La/B4C multilayers with a reflectivity of 40% in a 0.06 nm bandwidth near 6.7 nm.33 The theoretical reflectivity is close to 70%.34 Both Gd and Tb have 4p64dn4p54dn+1+4dn14f transitions near this wavelength and emit an intense UTA. The emission processes are more efficient than in Sn because the 4f wavefunction is completely contracted in the relevant ions, thereby maximizing the oscillator strength and leading to a better overlap of emission in adjacent stages than was obtained with Sn. In addition, transitions of the type 4d104fn4d94fn+1, not present in Sn, may also contribute to the in-band emission. Thus many ion stages can contribute and strong emission is expected over a wide range of plasma temperatures. The spectra of vacuum spark and laser plasma emission from laser-produced plasmas of Gd and Tb were first compared by Churilov and co-workers.35 In modeling the optimum conditions, the presence of these open 4f subshell configurations make calculation of spectra much more complex than in the case of Sn, because of the greatly increased number of transitions possible and the close proximity of configurations containing variable numbers of 4f, 5p, and 5s electrons in lower ion stages.36 Indeed, it is virtually impossible to make complete detailed calculations for excitation and radiative processes for these elements using existing atomic structure codes. As a result, despite the similarity in many instances to Sn, because of the difficulties posed by the presence of open 4f subshell ions, no full hydrocode simulation has been reported to date and thus no predictions exist for the variation of CE with electron temperature. However, even in the absence of such a calculation, it is possible to make some reasonable assumptions. The spectra from plasmas where the Δn=0, n=4n=4 UTA contains contributions from both open 4f and open 4d subshell ions has been studied in detail for tungsten because of its importance as a wall material for fusion devices.37 In all reported studies the emission is dominated by Ag- and Pd-like lines, i.e., the spectra containing fewest lines where the emission is not divided among many transitions. Thus, for Gd, for example, the strongest lines should result from 4d4f lines in the spectra of Ag-like Gd XVIII, Pd-like Gd XIX, and Rh-like Gd XX and the optimum electron temperature in a Gd plasma should, most likely, correspond to that which maximizes the population of these ions. The strongest transitions from Ag- and Pd-like ions have been identified3840 and are listed in Table 1. Results calculated in the present study with the Cowan suite of codes41 as well as those from other theoretical calculations are also included for comparison. The configurations chosen in the present case are given in Table 2. The Slater Condon Fk, Gk, and Rk parameters were reduced by 15% while the spin orbit integrals were left unchanged. The transitions from Ag-, Pd-, and Rh-like ions around 6.7 nm are presented in Tables 30040050060078 for Gd and Tb, respectively. Since the optimum center wavelength is yet to be decided, a glance at this table would suggest 67.65 as a suitable choice.

Table Grahic Jump Location
Table 1Wavelengths of the strongest lines occurring in the spectra of Gd and Tb in the 6.x nm region.