2.1.

First Signs of EUVL

In the author’s view, the EUV source technology was always the most critical issue in realizing EUV lithography. While some people such as Silverman,⁷ Lin,⁸ and Levinson⁹ reviewed EUVL, the source technologies were only briefly described. They did not mention in detail the historical change from a Xe plasma to a Sn plasma.

As a person involved deeply in EUV source development and as the person who strongly advocated the necessity of using tin as a fuel, the author would like to review the history of EUVL with the emphasis on the EUV source technologies.

To the author’s knowledge, the first proposal of EUVL was made by Kinoshita¹⁰ in 1986. Later, they printed 500 nm half pitch patterns.¹¹ EUVL was recognized as promising after a report of 50 nm printing¹² performed at Bell Labs of AT&T in 1990, and Japanese researches¹³ followed them. At that time, multilayer technology was immature, and polishing technology of optics was too poor to apply for fabricating EUV optics. However, among many technologies to be developed for EUVL, the most difficult technology was an EUV source.

Exposures demonstrating feasibility of EUVL by Kinoshita,¹¹ Bell Labs,¹² and Japanese groups¹³ were performed using synchrotron radiation (SR). However, SR can never be a source for EUVL. Most people might think the reason is the size and cost of SR, but that is not true. The true reason is that SR cannot supply the EUV power required in EUVL. SR emits collimated light, and the EUV intensity on a sample is very high, but the power integrated over a large solid angle is very low, because SR emits light only within an extremely small solid-angle. In order to get EUV power of more than two orders of magnitude higher than available from SR, an ultra-high brilliant plasma point source had to be developed.

2.2.

Full-Scale Development of EUVL Initiated by Xe Gas Jet LPP

2.3.

Uniqueness of EUVL Development

The biggest difference of the way to develop EUVL technologies from the way in other lithographies is not related to vacuum but to the way that EUVL needed in terms of basic research. In KrF and ArF lithography, most of necessary technologies were already established, and a trial-and-error approach worked. In EUVL, the trial-and-error approach does not work to solve problems of debris generation, erosion, and so on. Deep knowledge of physics is required to solve fundamental problems which appeared during development of EUVL technologies, including the choice of a fuel for an EUV source.

2.4.

Change from a Xe Plasma to a Sn Plasma

2.4.1.

Advocacy of a Sn plasma

In the mid-1980s, O’Sullivan et al. found that a strong narrowband emission is generated from a plasma of rare earth elements, that the peak wavelength scales with the atomic number of the element, and that a Sn plasma emits a strong band with the spectral peak at 13.5 nm.¹⁷ While many people knew the works by O’Sullivan et al., no one considered a Sn plasma as an EUV source, because a Sn plate was notorious in generating tremendous amount of debris. Furthermore, the practical conversion efficiency (CE) of a Sn plasma produced on a Sn plate was not so high compared with other elements like gold.¹⁸

At the early stage of EUV source development, people discussed the plug efficiency only. The author emphasized the importance of reducing heat load in vacuum. The author argued that increasing CE and the collection efficiency were critically important to reduce heat load in vacuum, which would limit extractable EUV power. The author claimed that Sn is the unique choice for a fuel to increase CE, and that an LPP is the choice for realizing high collection efficiency. The author advocated the use of tin in the second EUV workshop in 2001¹⁹ as shown in Fig. 1, and in the 3rd EUVL workshop in Matsue in 2001²⁰ as shown in Fig. 2.

Fig. 1

The view graph used in the author’s talk at the second EUVL workshop in 2000¹⁹ to emphasize the importance of reducing heat load in vacuum.

Fig. 2

The view graph used in the author’s talk at the third EUVL symposium in 2001²⁰ to advocate that the Sn LPP is the unique solution for high-volume production EUVL.

3.1.

Theoretical Limit of EUV Power

The author mentioned²⁰ that any source cannot exceed the blackbody brilliance, $P_{\mathrm{BB}}$. Power of a radiation source into a solid angle of $2\pi \mathrm{sr}$ is given by the following equation²¹ when opacity of the source ${\tau }_{\mathrm{opa}}$ is well smaller than 1.

3.2.

Spectral Efficiency

As explained above, it was well known that a tin plasma emits a strong narrowband emission band at 13.5 nm from the works¹⁷ performed in the 1980s. However, the observed CE at 13 nm within 2.5% bandwidth for a tin plate was only 1%, and the difference between CE for a gold plate was small.¹⁸ Hence, the simple change of a Xe plasma to a Sn plasma does not increase CE so much. Therefore, we have to examine how CE is determined.

Total conversion efficiency ${\eta }_{\mathrm{total}}$ is given by a product of four efficiencies:²¹ energy deposition efficiency ${\eta }_{\mathrm{abs}}$, radiation efficiency ${\eta }_{\mathrm{rad}}$, spectral efficiency ${\eta }_S$, and collection efficiency ${\eta }_G$.

As shown in Fig. 3, the spectral profile of a plasma generated on a tin plate in usual irradiation conditions is very broad, and spectral efficiency ${\eta }_S$ is very low, which leads to a low CE.

Fig. 3

A spectrum from a plasma generated on a tin plate is broad, and spectral efficiency is low, causing a low CE. By using a cavity confined target, we succeeded in realizing high spectral efficiency first with an LPP.

The broad spectrum of a plasma on a tin plate is caused by a large opacity, ${\tau }_{\mathrm{opa}}$,the optical thickness at 13.5 nm. The brilliance of a source, $P$, is given by

By increasing the plasma thickness, the intensities of weak emission peaks, whose opacity is low, grow near linearly,$P\sim P_{\mathrm{BB}}\tau$, but the intensity of a strong peak, whose opacity is large, saturates to the blackbody brilliance, $P\sim P_{\mathrm{BB}}$. Hence, the ratio of the energy flowing to the 13.5-nm band is reduced. Therefore, for a high spectral efficiency, the opacity at 13.5 nm must not be too large.

In order to realize high spectral efficiency, the author devised the cavity confined scheme,²² in which a material for plasma generation is supplied by laser ablating a concave surface in order to reduce opacity²³. As seen in Fig. 3, we succeeded in demonstrating that high spectral efficiency can be achieved by a Sn plasma. The spectral efficiency, ${\eta }_S$, of the cavity confined plasma shown in Fig. 3 was estimated to be 10%. From spectra reported for DPPs, we know the spectral efficiency ${\eta }_S$ can be as high as 18.6%.²⁴

3.3.

Opacity and Plasma Density

When a plasma is very thin, brightness of the source is proportional to the opacity, and opacity is to be increased to increase EUV power. On the other hand, for a high CE, opacity must not be too large, as discussed above. The optimum opacity is considered to be in the range $\tau =0.2\sim 1.0$.

The absorption cross section, $\sigma$, of a tin plasma at the wavelength of $13\sim 14\mathrm{nm}$ is reported²⁵ to be $\sigma \sim 2\mathrm{E}-17{\mathrm{cm}}^2$. If this value is correct, when the plasma dimater is $D=500\mu m$, the electron density $n_e=4.2\mathrm{E}18{\mathrm{cm}}^{-3}$, and the averaged charge state of the plasma, $Z$, is 12, the opacity, $\tau$, of the plasma is calculated to be

With this near optimum opcaity value, the spectral efficiency, ${\eta }_S$, can be as high as 12%. Then, when the product of the absoprtion efficiecny and the radiation efficiency is 70%, total conversion efficiency, ${\eta }_{\text{total}}={\eta }_{\mathrm{abs}}\times {\eta }_{\mathrm{rad}}\times {\eta }_S$, can be as high as $4.2\%/2\pi \mathrm{sr}$. And the brilliance of the source becomes $4.2\mathrm{E}9\mathrm{W}/{\mathrm{cm}}^2\times 0.35=1.5\mathrm{E}9\mathrm{W}/{\mathrm{cm}}^2$.

3.4.

Radiation Efficiency

We showed theoretically and experimentally²⁶ that there exists the optimum electron density which gives the highest radiation efficiency.

When heat conduction is neglected, a laser absorbing layer having a thickness of $L_{\mathrm{abs}}=1/\alpha$ can be considered as the emitting layer. We assume a laser light is absorbed via inverse bremsstrahlung with the absorption coefficient of $\alpha$, which is given by²⁷ $\alpha =aZ{n}_e^2$, where $a=6E-37{\lambda }_{\text{laser}}{(\mu m)}^2/T_e{(\mathrm{eV})}^{3/2}{(\mathrm{cm})}^5$, ${\lambda }_{\text{laser}}$ is the laser wavelength, $T_e$ is the electron temperature, $Z$ is the charge state of the plasma, and $n_e$ is the electron density.

The laser absorbing layer expands with the expansion velocity $V_{\exp }$ given by²⁸

When the electron density is very low, the radiation efficiency increases for a higher density, because the photon flux per ion is proportional to the electron density. On the other hand, when the electron density is very high, the radiation efficiency decreases for a higher density, as explained below.

Because the absorption coefficient $\alpha$ is proportional to the square of the electron density, the laser absorbing layer becomes transparent to a laser light after the expansion time defined by $t_{\exp }=L_{\mathrm{abs}}/V_{\exp }$. Then, the next layer in the target absorbs the laser light and expands shortly. Thus, when a laser pulse duration $t_{\text{laser}}$ is longer than the expansion time $t_{\exp }$ of the laser absorbing layer, i.e., $V_{\exp }{t}_{\exp }=L_{\mathrm{abs}}<L_{\exp }=V_{\exp }{t}_{\text{laser}}$, a target is kept ablated during laser heating, and the total number of ions in the plasma $N_{i,\text{total}}$ increases with time,

The emission power of the layer is lost after expansion, because the radiation power per ion is proportional to the electron density, and the emitting volume and radiation power proportional to $N_{i,\mathrm{emit}}$ remain unchanged. Therefore, an increase of $N_{i,\text{total}}$ increases the energy for creating the plasma without an increase of radiation energy, and then the radiation efficiency decreases.

From these considerations, the maximum radiation efficiency will be achieved when $L_{\exp }=L_{\mathrm{abs}}$. This was experimentally confirmed²⁶ by using the particle-distributed target.

In the ideal plasma, the radius, $r$, should be close to the laser absorption length. Therefore, the optimum laser pulse duration, $t_{\text{laser},\mathrm{opt}}$, is given by

The plasma expansion velocity, $V{}_{\exp }$, is given by Eq. (8), $Z/M\sim 10$ and $T_e\sim 50\mathrm{eV}$, then $V_{\exp }\sim 3.5E6\mathrm{cm}/\text{sec}$. Then, when the diameter of a plasma, $D$, is 500 μm, $t_{\text{laser},\mathrm{opt}}\sim 7\mathrm{ns}$.

When the radiation efficiency is large, however, the energy for creating a plasma is very small compared to the radiation energy, and then the effect of increase of plasma creation energy is not large. Therefore, the condition discussed above is not a stringent one. It means that allowance for the laser duration will be very wide. The most stringent condition for achieving a large CE is the opacity.

3.5.

Upper Limit of Repetition Rate

There exists an upper limit in the repetition rate. This was one of the hot topics of the Xe jet target technology at the early stage of EUV source development. Inolite Inc. showed beautiful pictures which showed that a frozen xenon jet was bent by the plasma created on the jet.²⁹

A high-temperature plasma exerts a high pressure. As shown in Fig. 4, a jet of ${\mathrm{SnO}}_2$ suspension was blown off by the high pressure of a plasma.²¹ The bright spot in the picture is the visible emission from a plasma. If a droplet target is too close to a plasma, it is blocked to reach the position of laser focus by the pressure of the plasma created in the previous shot.

Fig. 4

The pressure by a plasma generated on a jet of ${\mathrm{SnO}}_2$ suspension of 10 wt% concentration exploded un-irradiated jet stream.²¹ This shows the repetition rate of LPP is limited by the delivery of targets.

As shown in Fig. 5, the next target droplet must travel through a strong wind of a plasma generated by the previous target. For simplicity, we assume that the plasma has a uniform density profile with the initial density and radius of $n_0$ and $r_0$, respectively, and expands uniformly. Then the momentum given to the next droplet target is given by $m{V}_{\exp }SL{n}_0{(r_0/L)}^3$, where $m$ and $V_{\exp }$ are the mass and expansion velocity of ions in the plasma, $S$ is the cross section of the travelling droplet, and $L$ is the separation of droplets. When all material of the target droplet is converted to a plasma, the mass $M$ of a droplet equals to $m$ $n_0(4\pi /3)r_0^3$. Therefore, the condition for a next droplet of diameter $d$ travelling in the plasma wind to reach the laser focus point without a significant delay, is given by

Fig. 5

The repetition rate of target delivery is limited by plasma pressure.

Hence, for the next target droplet to travel in the wind of the plasma generated by the previous droplet, the ratio of diameter to the separation of droplets should satisfy

When the diameter of targets, $d$, is 30 μm, the separation of targets, $L$, needs to be $L>1.03\mathrm{mm}$. Then the upper limit of the repetition rate, $R=V_{\mathrm{ej}}/L$,is calculated to be

When all the target material is not converted to a plasma and the large fraction of core of targets remains solid, the plasma pressure is low, and the maximum repetition rate can be higher than the value given by Eq. (11).

The repetition rate can be increased by decreasing the diameter of targets. However, an increase of the repetition rate decreases the EUV power as explained below.

The EUV energy in one shot, $E_{\mathrm{rad}}$, is proportional to a product of surface area, the opacity, and the emission duration of the source. When all the target material is not converted to a plasma, the opacity and the emission duration are proportional to the plasma diameter. The surface area is proportional to the square of the diameter. Then the EUV emission power increases with the fourth power of plasma diameter,

For stable generation of droplets, $d/L=2\sim 8$ and $V$ is proportional to $1/d$. Then,

Thus, in order to increase the EUV power, the diameter of the plasma should be increased if the laser pulse energy is large enough, even the repetition rate decreases.

3.6.

Single Pulse Laser Energy

When the conversion efficiency is CE, and when the diameter of a plasma is $D$, the opacity is $\tau (\tau <1)$, and the laser pulse duration is $t_{\text{laser}}$, the laser pulse energy, $E$, absorbed by a droplet is given by

As discussed in Sec. 3.5, in order to increase the EUV power, we need to increase the plasma diameter. If we need a plasma of diameter $D=500\mu m$, the laser pulse energy needs to be larger than $0.4J\times (500/200)=2.5J$, when other conditions are the same.

3.7.

Target Diameter and Plasma Diameter

If all material of a solid density target with a diameter, $d$, is converted to a plasma of a diameter, $D$, having a uniform density profile of ion density, $n_i$, the target diameter, $d$, is given by,

3.8.

The Ideal Target; Particle-Cluster Target

From the above discussion, we know the diameter of the target for a plasma of 200-μm diameter is only 4.5 μm. Very old classical pre-pulse technology is very often applied to create a low-density plasma having a long scale length on a solid density target. However, even with pre-pulse irradiation, it is impossible to generate a plasma of 200-μm diameter by using a solid density target of 4.5-μm diameter. As explained later in Sec. 4, the target diameter is as large as 30 μm in Cymer’s LPP, which means only 0.34% of the target mass is used.

As a method of generating a uniform density plasma with a large diameter by using a solid density material, the author invented the particle-cluster target scheme,^30,31 in which a cluster of fine particles is delivered to the laser focus point and constituent fine particles are dispersed by some shocks just before shooting the laser, as shown in Fig. 6. This is the ideal target scheme for an EUV source. The delivered mass is minimized, which minimizes contamination, and the plasma density profile is flat-top.

One method of delivering and creating particle-clusters is the use of a suspension including tin particles of tens-nm diameter. Droplets of the suspension can be generated at a multi-kHz repetition rate. By vaporizing a solvent of a droplet, fine particles included in the droplet form a particle-cluster with weak coupling. The author believes that the ultimate high CE is achieved and contamination of the plasma is ultimately minimized only with the particle-cluster scheme. However, it will take a few years to develop a few difficult technologies for implementing the scheme, and people may not be patient to wait for a few years.

Fig. 6

The ideal method of delivering target material to generate a uniform density plasma with a large diameter.³⁰

The author wishes LPP suppliers to improve their simple scheme of shooting a single droplet for preventing contamination of surrounding optics by the plasma.

3.9.

Controllable Parameters

Although we see four parameters controlling the EUV power in Eq. (1), $D$ is the only one free parameter. As explained above, the opacity ${\tau }_{\mathrm{opa}}$ should be in the range between 0.1 and 1, and the upper limits of pulse duration $t_p$ and repetition rate $R_{\mathrm{rep}}$ are determined by the diameter $D$ of a plasma. In order to increase $D$, the laser pulse energy $E_{\text{laser}}$ should be large.

3.10.

Shielding of a Plasma by a Magnetic Field

Because a plasma is composed of charged particles, ions, and electrons, a magnetic field should have some effect to stop a plasma. We performed an experiment to evaluate the stopping power of a magnetic field and reported the result in 2003.^32,33 In order to generate a uniform density plasma, a through-hole target²³ was employed. Faraday cups with positive voltage bias to the cup were placed at 5 cm from the plasma on the backside of the target, and we observed a plasma flowing through the hole of the target. A permanent magnet of 2-cm width was placed between the plasma and Faraday cups. The observed Faraday cup signal is shown in Fig. 7. With a magnet of 1400 gauss, the ion signal dropped to near $1/10$. When the magnetic field was increased to 5000 gauss, the integrated signal decreased by three orders of magnitude.

The observed decrease of the ion signal was not caused by Larmor motion of ions and electrons along a magnetic field. Density of a laser plasma is so high, separation of ions and electrons is of the order of nm. On the other hand, the Larmor radius is sub-mm and micrometer for ions and electrons, respectively. Therefore, it is clear that, in an LPP, even electrons cannot make Larmor motion, but electrons and ions behave as a fluid. An interpretation based on a static pressure also failed to explain the observed results. Magnetic pressure is given by $B^2/2$, and it is 4000 Pa for $B=0.1T$. The pressure of a laser plasma is given by $n_e{T}_e$ and is as high as 1 GPa initially. However, after plasma expansion, density and temperature decrease, and the plasma pressure drops rapidly with expansion. When a plasma of the initial diameter of 0.1 mm expands to 20 mm, the pressure decreases to several thousands Pa. Therefore, we can expect that a plasma will be reflected at a few cm from the target by a magnetic field of about 1000 gauss. Plasma should be totally blocked when the strength of a magnetic field is above the threshold value. The observed behavior was totally different from this simple expectation. This disagreement is reasonable because a plasma is a compressible fluid. When a plasma is stopped by a magnetic field pressure, it is recompressed, and the pressure increases. To predict the interaction precisely, we have to take into account three-dimensional dynamical behavior of a flowing plasma.

Fig. 7

Blocking of high energy ions by a magnetic field. While ion signal decreased by several orders of magnitude, it was concluded that perfect shielding was difficult in a conventional configuration.^32,33

Although the interaction of a plasma with a magnetic field is not simple, as discussed above, we showed a magnetic field has a power of stopping a plasma to some extent. If some clever configuration is invented, plasma shielding by several orders of magnitude or larger may be possible.

Because a magnetic field cannot stop neutrals, all material of targets needs to be ionized for a magnetic field scheme for stopping contamination to work.

4.1.

Sn LPP at Cymer and Gigaphoton

Cymer developed technologies of generating tin droplets with extremely high stability and accuracy.³⁴ Their success of making a tin LPP as a real product exceeded the author’s expectation several years ago, and their efforts so far should be evaluated.

They reported⁵ that they developed the high-volume manufacturing 1 (HVM1) system in which a ${\mathrm{CO}}_2$ laser system having the capability of 30 kW maximum power irradiates tin droplets to generate 13.5 nm EUV light. They reported that with the HVM1, the IF equivalent EUV power of 11 W was observed when the laser was operated at 50% duty cycle.⁵

The diameter of tin droplets generated at 40 kHz was 30 μm,⁵ and the ejection speed of droplets was $30\mathrm{m}/\mathrm{s}$.³⁵ The pulse duration of the ${\mathrm{CO}}_2$ laser is about 15 ns,³⁶ and the focal spot size of the laser is about 150 μm.³⁵

The IF equivalent power of 11 W was calculated⁵ by assuming 5 sr for the collection solid angle, 50% for the reflectivity of a collector mirror, 90% for the transmission of buffer ${\mathrm{H}}_2$ gas for suppressing contamination of the collector mirror by the plasma, and the transmission of 65% for a spectral purifying filter. Then the power at the source is calculated to be

The power of ${\mathrm{CO}}_2$ laser in the experiment was not explicitly described in the report. If we assume it was 10 kW, CE into $2\pi \mathrm{sr}$ is calculated to be only 0.47%.

Cymer also reported that, by using the LT1 system, they observed the IF equivalent power of 160 W with the ${\mathrm{CO}}_2$ laser power of 17.5 kW with a pre-pulse with the estimated CE of 3%.⁵ The duty cycle of the laser irradiation in this case was as low as 3%.

The big difference of 11 W and 160 W in HVM1 and LT1 experiments might be attributed to the difference of duty cycle. They mentioned distortion of ${\mathrm{CO}}_2$ laser optics for a high duty cycle operation. Therefore, the apparent very low CE when 11 W at IF was observed may have been caused by 50% duty cycle operation of the laser. Real laser power irradiated tin droplets may have been very low due to distortion of optics.

With hydrogen buffer gas, they reported no degradation of reflectivity of the collector mirror was noticed while delivering a dose of larger than 4 mega joule to the IF, which corresponds to 512 wafers for $10\mathrm{mJ}/{\mathrm{cm}}^2$ resist sensitivity.⁵ By considering the difficulty of suppressing Sn contamination, this is a great achievement, but this lifetime is still very short, because it is only five hours for throughput of 100 wph. One year has 8,766 hours.

The other LPP supplier, Gigaphoton, also uses a ${\mathrm{CO}}_2$ laser for shooting tin droplets. They claim that the clean EUV power of 20 W was calculated to be achieved at IF with their ETS device when tin droplets of 30 μm diameter were shot by a 100 kHz ${\mathrm{CO}}_2$ laser of 3.6 kW power at duty cycle of 5%. They reported that the system was operated continuously about seven hours. The calculated CE was 2.1%.

They claim use of a magnetic field is their originality. They reported fast ions across magnetic field was reduced greatly.

4.2.

Interpretation of the Performances of Present LPPs based on the Author’s Theory

Here the author interprets the present status of sources based on the author’s theory.

4.3.

Prospects