High-volume manufacturing requirements drive EUV source development
05/01/2005
Extreme-ultraviolet (EUV) light source development has regularly been characterized as one of the critical issues facing the viability of EUV lithography. Output power and operational lifetimes remain top concerns for EUV sources. Industry efforts to date have focused primarily on the technical feasibility of one approach or another, but less emphasis has been placed on the critical issue of commercial feasibility to reach high-volume manufacturing (HVM). This article summarizes the technical viability of discharge-produced plasma (DPP) and laser-produced plasma (LPP) light source concepts when compared to HVM requirements, and also describes Cymer’s new LPP source development roadmap, which supports introduction of EUV lithography for the 32nm process node.
EUV light is generated by the deposition of laser or electrical energy into a source element, such as xenon (Xe), tin (Sn), or lithium (Li), creating ionized gas microplasma at electron temperatures of several tens of electron volts. As these highly excited ions decay, energetic radiation is emitted in all directions. For EUV lithography, the 13.5nm radiation is collected by a mirror (either grazing incidence or normal incidence) and focused to an intermediate point where it is relayed to the scanner optics and, ultimately, the wafer.
Cymer historically has pursued a dense plasma focus (DPF) DPP EUV source using Xe as the source material because of its simplicity and overall electrical efficiency. Early in 2004, the company began researching a variety of alternative source elements as well as LPP source technologies to support the introduction of EUV lithography for IC production at the 32nm node by 2009. Due to their configuration, DPP sources face extreme electrode heat-extraction challenges. Over the years, the company focused on improving electrode thermal extraction from smaller electrodes, and in 2004, the company demonstrated the ability to extract more than 20kW from source electrodes. Such operational power levels, however, are below those required to reach HVM source powers. These levels are also at the extreme limits of what can be accomplished with known material and cooling technologies. Additionally, large source sizes do not allow source spatial combination as a scaling technique, and the mitigation of large amounts of electrode debris remains challenging.
In fall 2004, the company decided that the best path to an HVM-capable EUV source was through the use of a novel LPP concept. This decision was based upon the expectation that the requirements for the ultimate HVM products will be in excess of those predicted today and far beyond the capability of DPF technology concepts.
The new LPP approach
This concept uses Li as the source material and an excimer-based drive laser, shown schematically in Fig. 1. Li has been identified as an ideal source element due to its high conversion efficiency (CE) and its unique ability - among condensable source elements - to be evaporated from the surface of the collector mirror at relatively low temperatures.
 Figure 1. LPP concept uses Li as the source material and an excimer-based drive laser. |
The drive laser is a frequency-tripled Nd:YLF master oscillator driving two xenon fluoride (XeF) excimer power amplifiers firing in a time-interleaved fashion. This configuration combines the solid-state laser benefits of high-frequency operation and excellent beam quality with the lower cost and reliability of gas-discharge excimer power amplifiers.
 Figure 2. Conceptual drawing of an HVM EUV source system using two drive lasers in a sub-fab configuration. |
Figure 2 depicts the concept of an HVM system using two drive lasers located in a sub-fab configuration. A beam transport system is used to deliver the laser outputs to the source chamber. It compensates for slow drifts between each of the drive laser amplifier chains and the target system. The beam transport system also allows two or more drive lasers to be directed to the source chamber. The source chamber contains a heated, 5-steradian, normal-incidence mirror to collect the EUV emission and direct it to the lithography tool. It also contains a Li target delivery and recovery system, collector-mirror protection system, and supporting metrology systems. The source chamber will be embedded within the lithography tool, as it must be highly integrated with optical and vacuum systems.
The source chamber
Figure 3 plots CE measurements for Li as a function of intensity for a variety of drive laser wavelengths. Figure 3a is a comparison of these measurements to modeling results. The general shape shows good agreement between the model and laboratory CE-vs.-wavelength measurements, with the absolute measurements exceeding the model calculations. Even in this nonoptimized condition Li’s CE is high, approaching 3% and relatively insensitive to laser wavelengths from 266nm-1.064µm. This CE is similar to that reported for Sn and much higher than the CE for Xe.
 Figure 3. Comparison of a) measured and b) calculated CE results at 13.5nm. |
null
 Figure 4. Photo of 100µm droplets at 36kHz. |
Li has a very narrow spectral output peaked at 13.5nm. The Li spectrum has been measured to be significantly narrower than either Xe or Sn. This strong emission line is centered in the reflectivity curve for the planned multilayer mirrors (MLM) and allows for realistic variation in the central wavelength - unlike Xe or Sn - for the numerous mirrors within the scanner’s optical system without additional loss of EUV power. Measurements comparing the spectral output from Sn and Li in the EUV range and the UV-visible band from 200-800nm show that Li has much lower integrated spectral output in these bands, possibly eliminating the need for spectral purity filters.
A liquid metal droplet generator, compatible with both Li and Sn, was developed that can produce stable droplets <100µm dia. at repetition rates up to 48kHz with a working distance of 50mm. Figure 4 shows such droplets being illuminated by the XeF laser.
Collector protection
The EUV community has begun to focus additional efforts on developing ways to protect to the collector mirror, one of the single most costly elements in the EUV source. Its performance can be degraded by deposition of the source element and debris; diffusion of these materials into the MLM structure; and fast ion sputtering of the MLM coating.
One of the key advantages to using Li as a source material is that it is possible to minimize the amount of Li deposited on the collector surface by maintaining the collector at an elevated temperature. Thus, the evaporation rate of Li from the collector surface is significantly higher than the influx rate of Li from vaporized Li droplets to the collector. Given the droplet sizes and geometries, we expect that the evaporation rate would equal the influx rate at slightly >350°C and that a temperature of 400°C would cause the evaporation rate to be 10× the influx rate.
To heat the collector to 400°C, a MLM coating must be developed that remains stable for thousands of hours at these elevated temperatures. Several coatings have been developed over the past year, with reflectivities of nearly 70% that remain stable at 500°C.
Li diffuses through many materials at elevated temperatures, including standard MoSi MLM structures. An investigation of high-temperature materials as diffusion barriers to incorporate into the MLM coating structure has identified several promising candidates. Analysis by secondary ion mass spectroscopy (SIMS) demonstrated the effectiveness of one of these barrier materials (Fig. 5). The data show that Li does not significantly diffuse into the MLM and that the periodic MLM structure remains stable after exposure at 400°C.
 Figure 5. SIMS analysis of MLM structure. |
Another problem that affects the lifetime of the primary collector mirror is sputtering of the MLM coating by high-energy ions generated in the plasma. The physical removal of coating material degrades mirror reflectivity, reducing potential throughput of the lithography tool. In addition, any nonuniform removal of the coating leads to the nonuniform angular distribution of EUV radiation, reducing uniformity of mask illumination and potential yield of ICs. In experiments, energies of ions produced in laser plasma reach several kilo-electron volts. Without collector protection, mirror lifetime would be limited to several million pulses. At expected repetition rates of 10kHz or more, this would correspond to just minutes of production time. Collector mirror lifetimes of >100 billion pulses are believed to be required for commercial viability. Several fast ion-energy mitigation techniques are under development, including the use of buffer gases, foil traps, and electric and magnetic fields.
 Figure 6. Faraday cup signals with and without debris mitigation. |
In the experiment, energies of Sn ions produced in an LPP source were measured with a Faraday cup detector. Figure 6 represents the Faraday cup signal for two cases: without debris protection (black line) and with debris protection (red line). The signal vs. time represents the intensity of ion flux vs. ion energy. The fastest ions in this experiment have energies of 5keV and the integral under the curve represents the total ion flux for one laser pulse. The results indicate a reduction in both peak ion energy and total ion flux. For this case, modeling predicts that sputtering of the MLM coating will be reduced by more than a factor of 10. Another significant advantage of Li is that at optimum CE laser intensities, the ion energies are 5× lower than for Xe or Sn and, when combined with the sputter yield for Li, result in a 10× reduction of sputtering of MLM coatings. Another factor of 10 can be achieved by using several sacrificial coating layers on the MLM. We expect a combination of these improvements will enable collector-mirror lifetimes into the tens of billions of pulses, which is acceptable for the first HVM tools.
Drive laser
Fortunately, the 351nm wavelength of a XeF excimer laser is within Li’s broad CE-vs.-wavelength curve. Cymer has developed a hybrid laser concept where the beam quality and high repetition rate are provided by a tripled Nd:YLF solid-state master oscillator, and the required pulse energy is provided by XeF power amplifiers. Novel turning techniques are used to shift the output wavelength of the tripled Nd:YLF sufficiently to match the XeF gain spectrum.
In the case of XeF lasers, the emission centered near 351.15nm results from a transition between the upper-level vibrational mode 1 to the lower-level vibrational mode 4. The transitions between vibrational levels 0 and 2 produce the XeF emission centered near 351.25nm. Gain extracted from the XeF gas can be stimulated by wavelength tuning an Nd:YLF laser to match this 1-to-4 XeF transition.
Extensions to Cymer’s existing excimer technology have demonstrated 800W operation with the required beam divergence and repetition rates. A development laser, dedicated to material and laser component life testing, has been tested for more than 9 billion shots with no observed optics degradation. We expect that higher-capacity pulse power and improved discharge chamber technologies will achieve 2.3kW at 12kHz by the end of this year, and 3.5kW/laser at 16kHz in 2007.
Industry requirements
The current joint requirements for HVM include an EUV source power of 115W at intermediate focus (IF), as shown in the table, with an operational lifetime of 10,000 hr. It is important to note that the power requirement assumes that a spectral purity filter is not required and a suitable 5mJ/cm2 EUV photoresist can be developed.
 |
One of the significant challenges for any EUV source to be commercially viable is affordability. For an LPP source to be viable, it must meet these requirements:
- Selected source material must have a high CE (≥4%) and be delivered in a way that minimizes the amount of debris generated.
- The collector must have a high average reflectivity, including losses at larger incidence angles of a 5-steradian collector. It must exhibit long life and be easily exchanged at a reasonable cost.
- The selected drive-laser technology must meet initial cost and cost of consumable goals while meeting the performance requirements for high CE.
- Laser power must be delivered efficiently to the source chamber through a beam transport system from remote lasers.
Over the past several years, substantial progress has been made on both LPP and DPP technologies. But what happens if the source power requirements increase to support photoresists that require 10 or 20mJ/cm2 or to provide increased process latitude or to meet the requirements of future scanners? What if a spectral purity filter is required?
The industry should keep in mind that deep-ultraviolet (DUV) photolithography history shows that the power required from the laser source did not decrease over time. On the contrary, from the time DUV excimer lasers were first used in a production environment, the power requirements actually increased sixfold. A viable EUV source technology must take into account that the power requirements for the ultimate EUV sources will likely exceed the 115W currently projected for HVM applications. Given these unknowns, a viable concept must offer the required scalability to support the industry.
EUV source power requirements were introduced at only a few watts in 1996 and have increased to over 100W at IF to compensate for collection efficiencies and enable a throughput >100 wafers/hr. We expect this trend to continue after the introduction of the technology to allow for low-sensitivity resists, process latitude, optics lifetime degradation, and general technology evolution in a competitive marketplace. LPP offers the highest degree of power scalability to meet future industry requirements.
Future developments
By year’s end, Cymer will have a system that can produce an equivalent of 15W EUV power at the IF using a single 2300W drive laser. Over the next few years, improvements in laser performance are expected in energy/pulse and repetition rate, ultimately resulting in a single laser that can produce 3500W. The combination of two such lasers in a system will produce the necessary 7000W of laser power needed to provide the >100W of EUV power at the IF required of an HVM source.
Conclusion
Li droplets combined with an excimer laser provide an optimal path to a EUV HVM source. An LPP-based system can meet the anticipated HVM requirements. The use of Li as the source material maximizes in-band CE and allows manageable debris mitigation through evaporation and lower source ion energies. Excimer lasers are today’s best drive laser solution and have a proven semiconductor-fab track record with affordable cost of operation. The primary development challenge for a viable source technology continues to be the development of debris mitigation techniques and MLM coating technology, to provide the necessary collector lifetime and cost for critical-dimension lithography production requirements.
Contact David W. Myers at Cymer Inc., 17075 Thornmint Ct., San Diego, CA 92127; ph 858/385-7300, e-mail [email protected].
Igor V. Fomenkov, William Partlo, David C. Brandt, Brian C. Klene, Cymer Inc., San Diego, California