B. Nikolaus, W.N. Partlo, I.V. Fomenkov, Cymer Inc., San Diego, California
Optical lithography will be replaced by a nonoptical technique at some point for device shrinks to 35nm. Extreme ultraviolet lithography is currently showing the most promise. However, issues such as power scaling and cost of consumables, necessary for a production-level source, are still works in progress. Cymer is proposing a concept for a 13.5nm source based on dense plasma focus to meet these challenges.
Performance increases and productivity gains through device shrinks have been the main drivers in IC manufacturing for the last 30 years. This trend is unlikely to change until the functional limits of conventional logic devices are reached at ~35nm. This is expected by the end of the decade.
When analyzing imaging requirements of future lithographic tools, the equations for critical dimension (CD) and depth of focus (DOF) shown in Fig. 1 provide a guide to some design requirements. For example, CDs can be reduced by decreasing the illumination wavelength or increasing the numerical aperture (NA) of the imaging lens. Increasing NA, however, reduces DOF and with it, process control.
Therefore, the shorter-wavelength solution may be preferable. Further, pushing process factor k1 below 0.5 induces nonlinear printing effects that require reticle enhancement techniques (RET) to ensure the fidelity of the printed pattern. This, in turn, drives lithography costs up.
By decreasing the illumination wavelength, smaller linewidths can be achieved without sacrificing DOF. Of the nonoptical techniques, extreme ultraviolet (EUV) is considered by many to be the lowest-risk path to meeting the necessary lithography requirements. An EUV illumination wavelength below 15nm with a 0.2 NA projection system can print 50nm lines with a DOF above 400nm.
Figure 1. DOF and CD vs. NA for 10nm and 15nm illumination wavelengths; a minimum CD is given at DOF = 400nm (k1 = 0.7, k2 = 1).
Many challenges accompany ease of imaging, however. EUV can affect overall system engineering. For example: sources and scanners have to work in a high-vacuum environment, or in very low pressure gases, like helium or argon, due to strong gas absorption in the EUV spectral range; all optics collectors, illuminators, projection optics, and mask designs must be made of reflective elements; no transmissive optical material exists for wavelengths <20nm; and minimizing optical coupling loss requires full integration of the source into the scanner tool.
Despite these challenges, EUV is strongly supported for the 50nm technology node. One area of concentration has been the EUV source, emitting in the 10-20nm region. Several source concepts have been developed in recent years [1-7] employing either laser or electrical excitation schemes. In either case, strong EUV emission is generated in a hot plasma of highly ionized xenon (Xe) gas (lines at 11.3nm and 13.5nm) or lithium (Li) vapor (strong line at 13.5nm).
Mirror technology for projection optics will influence selection of the first EUV tools. Using multilayer coatings of molybdenum/beryllium (Mo/Be), which reflect wavelengths between 11.2-12nm, or molybdenum/silicon (Mo/Si), between 12.8-14nm, has demonstrated ~70% peak reflectivity. Reluctance to use beryllium due to its known high toxicity, however, plus the possibility of a powerful Li source have favored the use of the 13.5nm illumination wavelength for the first tools.
Defining EUV source requirements
The source requirements can be defined as derivatives of tool performance criteria.
The top three requirements are:
- resolution and stability for image quality (CD control);
- high optical power at the wafer level for high wafer throughput; and
- long component lifetimes and low cost of consumables for low cost of operation.
In a step-and-scan exposure tool, a uniformly illuminated slit at the mask level is imaged onto the wafer plane with a typical reduction factor of four. With mask and wafer moving synchronously at speeds that depend on resist sensitivity, exposure field size, and source repetition rate, the whole mask pattern is imaged onto the wafer. Maximum tool performance is limited by stage accuracy and source stability, which leads to dose error and results in CD variations. To reduce the exposure error, the number of pulses superimposed within the scanning exposure field can be increased. That can be accomplished by either increasing the field exposure width or operating the source at a higher repetition rate.
The table summarizes some of the key parameters anticipated for the first-generation EUV scanner. Applying typical step and overhead times of current 300mm step-and-scan tools, a 6kHz/30W EUV source would be able to support throughput of 75wph. For high-volume production tools, most analyses call for EUV powers in the 50-100W range.
An EUV source concept
Dense plasma focus (DPF) devices have been investigated as sources for EUV lithography because of the potential for high source brightness and repetition rate operation. The electrical driver circuits are similar to those used in multi-kilohertz lithography excimer lasers, and proven and production-worthy solid-state pulse power technology is well suited for the 5-10J stored energy requirements of this DPF.
Figure 2. In the illustration of DPF, electrical and magnetic forces lead to a hot plasma pinch, the point source of intense EUV radiation.
An EUV source solution consists of DPF and collector optics, with an output rated by the amount of energy radiated into a Mo/Si reflective band centered at 13.5nm and multiplied by the geometrical and optical efficiency of the collector. Critical success factors in building an EUV lithography source for production are:
- high in-band conversion efficiency;
- efficient thermal cooling for operation at high average powers;
- reduction of the electrode erosion rate; and
- debris mitigation at the collector system.
It is apparent that the cost of operation (CoOp) will, to a large degree, depend on the lifetime of the collimating optics. Making the collimator a low-cost consumable as well as various mitigation strategies to prevent degradation of the collimator through contamination and debris are under evaluation. Though it is premature to publish CoOp estimates, it is worth noting that this DPF scheme has been favorably compared to competing technologies such as laser-produced plasma  and capillary discharge .
The operating principle of DPF is illustrated in Fig. 2. A plasma sheet is initiated between the inner and outer electrode at the base of a coaxial gas discharge configuration. As the current increases, the sheet is accelerated away toward the end of the electrode, where the gas is compressed by magnetic forces directed along the central axis. Due to compression heating, the plasma temperature can reach levels sufficient for intense emission of EUV radiation.
Initial research focused on a DPF that produced plasma temperatures appropriate for the 13.5nm emission line of doubly ionized lithium [3, 4]. Early characterization revealed that significant amounts of EUV radiation could be produced. For example, an early prototype converted 25J of stored electrical energy into 0.76J of in-band 13.5nm radiation into 4p steradians (str), which corresponds to a conversion efficiency of 3%. At 200Hz, the measured pulse-to-pulse stability was 6% (1s) with no decrease in conversion efficiency, representing an in-band EUV power of 152W radiated into 4p str.
The major drawback found in this prototype was poor position stability leading to a large integrated source size and high electrode erosion. Its scalability to multi-kilohertz repetition rates needed for better dose control was limited due to the high-energy input. Modifications, however, have resulted in greatly improved stability and much lower input energy requirements. Stable operation could be achieved with as little as 3J of stored electrical energy. Source size and stability were found to be acceptable for pulse repetition rates up to the power supply limit of 2500Hz. Further, magnetic compression stages similar to those used in excimer lasers were inserted, thus increasing the peak power flow into the final stage capacitor and enabling the delivery of 10.0J of energy to the DPF electrodes. The overshoot energy could also be fully recovered for the next pulse.
To use other source gases, the development of lithium handling and delivery was decoupled from the development of basic DPF performance in a fourth-generation machine. This system produces significantly higher plasma temperatures, allowing efficient EUV emission using a xenon (Xe) gas. With Xe as a source gas, the current-generation DPF exhibits efficiencies similar to other sources employing direct electrical drive of the xenon plasma. An emission efficiency of greater than 0.20% into 2p str at 2% bandwidth, centered at 13.5nm, has been measured.
Due to electrode erosion, the success of this source concept depends on the fabrication of grazing incidence optics, a collector element more tolerant to debris than multilayer dielectric mirrors. Mitigating debris by inserting a foil trap is also being considered. In either case, no significant impact on the source brightness is expected.
Lithium and xenon sources
The measured Li and Xe spectra, along with the published reflectivity of an EUV Mo/Si multilayer dielectric mirror, are shown in Figs. 3 and 4. The resolution of the spectrometer is sufficient to resolve many of the lines that make up the cluster of Xe+XI emission near 13.5nm.
We took Xe spectral scans with two different input energies to observe the relative impact on each of the emission clusters. Changing the input energy by 35%, from 7.7-10.4J, increased the emission from the xenon IX lines by only 30%, but the emission from the xenon XI lines increased by 220%. This result leads us to believe that we do not exceed the efficient plasma temperature for 13.5nm emission with xenon, even with the highest input energy available from this DPF.
In-band source brightness
A measurement vessel, based on the "flying circus" concept , was constructed to make measurements of the in-band EUV emission with xenon source gas. After confirming the near isotropic radiation characteristics of the source, the spherical expansion factor relative to 2p str for this arrangement was determined.
Figure 3. The Li+II 13.5nm line superimposed on the Mo/Si reflectivity function indicates the viability of using Mo/Si multilayer coatings for projection optics.
Absolute energy calibration of the Xe source includes a reduction factor of 2 to account for the reduced transmission efficiency of the Xe spectrum (compared with the 13.5nm single-line Li spectrum) in a 9 Mo/Si mirror-imaging system. This is the usable in-band Xe source emission otherwise noted as 2% bandwidth. Using this setup, we measured the emission into 2p str vs. input energy with xenon as a source gas. The resulting energy output and efficiency curves are shown in Fig. 5. The most notable result obtained from this data is that the efficiency continues to rise even at the maximum input energy, so it is unclear how much more improvement is available at higher input energy. The highest efficiency to date for this machine is 0.23% measured at 10.5J input energy.
Energy stability measurements
An important performance parameter for any EUV lithography source is emission stability. All EUV lithography tools will be scanner-based, and the slit width of an all-reflective projection optic is expected to be narrower than that for a conventional all-refractive projection optic. A narrow scanning slit allows only a limited number of pulses/subexposure region. The current requirement for energy stability is 3σ<2%.
Figure 4. The matching of DPF line spectra of Xe+XI with one of the Mo/Si reflectivity peaks suggests the possibility of Mo/Si coatings for xenon EUV systems.
Using the setup described in the previous section, the emission stability of the DPF was measured at low-repetition rate and high-repetition rate. The low-repetition rate data resulted in 1s = 9%, which is very far from the stated requirements. To better understand the nature of instability, the emission was measured at 1000Hz and compared with the 20Hz data. Again, the measured data for 100 pulse bursts at 1000Hz showed a standard deviation for energy variation of 9%. The result suggests that the mechanism for stability degradation does not depend on repetition rate. More research is necessary, however, to improve this source parameter.
Out-of-band emission measurements
Another important consideration for a production EUV source is the out-of-band emission. Since the Mo/Si mirrors also exhibit reasonable reflectivity in the UV/Visible (UV/Vis) region, energy in this wavelength range can make it to the wafer plane and degrade the aerial image. To eliminate radiation from this spectral region, a spectral purity filter (SPF) has been proposed. This filter, consisting of a thin membrane made of silicon or zirconium, would have an extinction coefficient of many orders for UV/Vis, but only 50% for 13.5nm radiation.
Figure 5. The resulting energy output and efficiency curves with the Xe source indicate a continuous rise in efficiency, even at the maximum input energy; therefore, higher efficiencies are possible.
To measure the out-of-band radiation, different filters were placed in the beam path, to select different spectral regions for evaluation. The results for these measurements are as follows: emission into 2p str, all wavelengths: 211mJ (2.0% of input energy); emission into 2p str, 11-20nm band (Be filter): 110mJ (1.0% of input energy); and emission into 2p str, 130-1300nm band (CaF2 filter): 0.8mJ (0.38% of all radiation).
Surprisingly, only 2% of the input energy is converted to radiation from the pinch. This means there is potential for significant in-band efficiency improvements once the mechanisms for the other 98% energy consumption are identified.
There is a very small fraction of radiation emitted into the UV/Vis region. This may allow the elimination of the SPF, thus avoiding the extra 50% loss in EUV.
EUV source size
Because EUV lithography tools will employ low-NA projection optics, the source must exhibit high brightness. If the size of the source is too large, then the allowed collection angle will be limited.
Figure 6. Off-axis images of the Xe source show an elongated image at 30µ as the camera is moved further off-axis.
A pinhole camera arrangement was used to make EUV images of the DPF with xenon as the source gas. This arrangement had an adjustable viewing angle between on-axis and 30µ off-axis. The resulting source images for several angles are shown in Fig. 6. These images are an integration of 1000 pulses to average out any source position instabilities. As the camera is moved further off-axis, the source image becomes elongated to the point where, at 30µ, the source is 250µmx1700µm FWHM.
Grazing incidence collector
Creating an EUV source with high spectral brightness is as important as collecting and relaying this radiation to a region safe for use with the sensitive Mo/Si EUV mirrors. Electrode erosion of the DPF source produces debris that can degrade Mo/Si multilayer EUV mirrors on direct exposure. Alternative collection optics is required. A multishell collector based on grazing incidence reflections offers promise, because it exhibits the necessary high reflectivity and can be fabricated from materials that are less sensitive to debris. In addition, if the design of the grazing incidence collector is made simple enough, it can become a low-cost replacement part.
The following performance parameters have been determined for the collector design: geometrical collection: 28.6% of 2p (two shells); and overall collection efficiency (including reflectivity losses): 18.6% of 2p (two shells).
Overall, our results and rate of progress with a DPF source have been encouraging and on schedule for providing a production-ready EUV source by 2005. The current xenon DPF prototype exhibits conversion efficiencies comparable to other direct electrical excitation sources. The quality of the grazing incidence collection optics has been improved, so that it no longer degrades the source brightness. The out-of-band UV/Vis radiation produced by this DPF source is very low and may allow the elimination of the SPF.
Energy stability, however, needs substantial improvement; power scaling requires considerable thermal re-engineering of the present design; and electrode erosion will require further investigation, including debris mitigation schemes to keep cost of consumables in an acceptable range.
At this point, a 13.5nm R&D source, based on DPF technology with xenon gas, could be built with the following parameters: thermally limited input power: 25kW; conversion efficiency into 2.0% bandwidth, 2p str: 0.20%; overall collection efficiency (of 2p): 18.6%; and maximum collectable EUV power at 13.5nm: 9.3W.
- R.H. Stulen et al., "Progress in the Development of EUV Lithography," SPIE 3676 Symposium on Microlithography, 1999.
- S. Okasaki, "EUV Program in Japan," SPIE 3676 Symposium on Microlithography, 1999.
- W. Partlo, I. Fomenkov, D. Birx, "EUV (13.5nm) Light Generation Using a Dense Plasma Focus Device," SPIE Proc. on Emerging Lithographic Technologies III, Vol. 3676, pp. 846-858, March 1999.
- W.T. Silfvast, "Intense Xenon Capillary Discharge EUV Source in the 10-16nm Wavelength Region," Opt. Lett. 23, pp. 1609-1611, 1998.
- W. Partlo et al., "Development of an EUV (13.5nm) Light Source Employing a Dense Plasma Focus in Lithium Vapor," Proc. of SPIE, Vol. 3997, Feb. 2000.
- F. Jin, M. Richardson, "Conversion Efficiency and Debris Studies of Ice Targets for EUV Projection Lithography," OSA Proceedings on Extreme Lithography, Vol. 23, No. 260, 1995.
- M. McGeoch, "Radio-frequency-preionized Xenon Z-pinch Source for Extreme Ultraviolet Lithography," Applied Optics, Vol. 37, No. 9, March 1998.
- R. Stuik et al., "Flying Circus EUV Source Comparison," presented at the EUV Source Workshop sponsored by Sematech, Nov. 2000.
For more information, contact the authors at Cymer Inc., 16750 Via Del Campo Court, San Diego, CA 92127; ph 858/385-7300; fax 858/385-7100.