Fast-Switching Valves for High-Productivity ALD
William Glime, Swagelok Co., Solon, Ohio; Tom Seidel, Genus Inc., Sunnyvale, Calif. -- Semiconductor International, 9/1/2005
|
High-productivity atomic layer deposition (ALD) processes are being driven by semiconductor applications (e.g., capacitor, gate and interconnects) requiring ultrathin films or conformal coatings with precise thickness control. ALD is particularly effective on surfaces with high aspect ratios or where graded compositions (i.e., one layer consists of one material and the next layer of another) are required. Future memory devices referenced in the International Technology Roadmap for Semiconductors (ITRS), such as magnetic RAM (MRAM) and phase-change memories, nanofloating gates, single-electron and molecular memories, may initially employ topologies with relatively relaxed aspect ratios compared with today's DRAM devices. But they all require ultrathin films, and eventually they will all migrate toward moderate, if not extreme, aspect ratios. The ITRS projects that DRAM capacitor deep trenches made ca. 2010 will require step coverage on high aspect ratios approaching 100:1 at 45 nm linewidths. The active area would be 20× the planar silicon. The challenge to provide conformal coating on such high-density structures will depend on significant advances over conventional ALD, including advances in ALD chemical precursors, delivery systems, and operating systems.1,2
Although ALD has already been used to make semiconductor and thin-film head devices in commercial production, future applications require improved productivity. New high-productivity modes of ALD, which provide a film deposition rate as much as 10× that of conventional ALD, show great potential for meeting the device requirements identified in industry roadmaps. These processes place new demands on the tools and components employed, including gas delivery systems, pumps, and fast-switching valves. In particular, valves play a critical role because they are the principal component responsible for controlling the flow of precursors, and they also purge gases into the chamber in precisely metered and timed pulses. Today's mass flow controller technology is not compatible with the short and frequent gas pulses inherent to high-productivity ALD processes. As a result, even greater reliance is placed on the precision and accuracy of valves used on ALD tools.
High-productivity ALDTo clearly understand the demands placed on fast-switching valves, we will review a sample high-productivity mode of ALD, limited optimized reactions by ALD (LORA), described by Kim et al.3 As in conventional ALD, a four-stage cycle is employed with two chemical precursors separated by a purge cycle. These processes are repeated to build the film. The cycle time is defined as the sum of the exposure and removal periods:
Cycle Time = Exposure 1 + Removal 1 + Exposure 2 + Removal 2
where 1 and 2 represent the precursor chemicals from which the deposited film is derived. In the LORA mode, cycle time is reduced to its optimum point for any given chemistry. The objective is to optimize the exposure and purge processes, which entails reducing the exposure and purge times. The result is more layers (or cycles) in less time and, in turn, greater thickness in less time. Productivity is increased by focusing on thickness per unit time (Å/min) rather than thickness per cycle (Å/cycle).
Genus intentionally decreased the exposure time below the point of total saturation. In turn, the required purge time became less because there was less precursor to remove. As the exposure and purge times are reduced, the thickness per unit time continues to increase, and then ultimately decreases as the reaction becomes very starved.
This approach runs counter to conventional ALD, which is usually understood to be a robust process in which the reactant chemicals are allowed to reach a saturation point of 96%+. Typically, the surface is "overdosed." In ALD, chemical utilization or efficiency is very high during initial exposure, but very low in the saturation region. In the LORA mode, underdosing is used, and the film that is formed is of high quality with desirable stoichiometry and good electrical properties.3
Near 100% conformity has been demonstrated for aspect ratios of ~40:1. Film-thickness uniformity has been achieved at the ~1% level for 100-2000 Å depositions. The composition is substantially stoichiometric Al2O3 (RMS). The electrical properties show high dielectric breakdown fields >8 MV/cm. The total cycle time is ~0.5 sec for the maximum deposition rate per unit time. Other chemistries and films will be reported in the future.
We developed a case study for the maximization of the film growth rate for a TMA/H2O chemistry. In Figure 1 , Al2O3 growth rates are charted using TMA and H2O as a function of H2O exposure time.
LORA is one example of a class of new short-cycle-time ALD modes. Rapid ALD may be carried out, for example, by the LORA mode or in a time-phased multilevel flow, where the purge entrainment flow is increased during purges relative to the exposure flows to reduce the purge time.4 This mode and variants of it also result in cycle times as low as ~0.5 sec.5 Hence, reliable fast-switching valves should be looked at as a general emerging requirement.
New valve requirementsPrior to ALD entering the semiconductor processing arena, the valves used to control the flow of chemicals into processing tools were cycled at a relatively low frequency, typically no more than a few tens of seconds. Thus, 1 million valve cycles was well beyond the life expectancy of the tool on which the valves were employed. With the entry of ALD processes, ultrahigh-purity (UHP) diaphragm valves were challenged with new performance demands, such as:
- Millions to tens of millions of cycles per year.
- Fast actuation (open and closing times of <20 msec).
- Consistent flow performance.
- Compatibility with high temperatures and new corrosive precursors.
At the same time, UHP diaphragm valves were still being required to provide the cleanliness associated with semiconductor applications.
Designing an ALD-compatible valveSpringless diaphragm valves are state-of-the-art for UHP gas delivery systems. The operation of a diaphragm valve (Fig. 2 ) depends on the deformation (flexing) of a metal diaphragm, which is pressed against a seal to close the valve and then allowed to relax away from the seal to enable flow through the valve. One of the principal compromises, or balancing acts, associated with a metal diaphragm valve involves maximizing the flow through the valve while minimizing the stress in the flexing metal diaphragm. More diaphragm deflection produces higher valve flow, but also produces higher stresses in the metal diaphragm.
An SN curve (stress vs. number of cycles), illustrated in Figure 3 , describes the performance of a material under cyclic fatigue conditions. Using the SN curve for the diaphragm material, valve designers can determine the maximum level of stress that can be accommodated in the diaphragm in order to achieve a given cycle lifetime. When targeting a valve for an application that requires only a few thousand cycles, the valve design can allow for a higher diaphragm stress than would be acceptable if the desired service was 1 million cycles. The maximum level of stress a material can repeatedly experience without failure is known as its "fatigue limit." For a diaphragm valve to provide reliable service for many millions of cycles, its metal diaphragm must operate below the fatigue limit of the material from which it is constructed.
UHP diaphragm valve design relates to the interplay between diaphragm, seat, button, and valve body. The specific application conditions, such as pressure, temperature, flow, chemicals, etc., dictate the optimal valve design. Through research and relationships with key ALD developers, the range of conditions where ALD processes operate is established. Early and emerging ALD tools and processes did not permit the flow capacity of the valves to be compromised to achieve the requisite cycle life. Further, the basic size of existing UHP diaphragm valves cannot change.
The need for fast precursor pulse delivery prompted the relocation of control valves from a remote gas panel to the top of the deposition chamber, where real estate is at a premium. Many systems were also already designed around standard valve products, so a significant change in valve dimensions would require tool redesign in some cases.
Given these constraints, the shape of the diaphragm becomes a focal point for achieving the desired order of magnitude improvement in valve cycle life. Typical springless diaphragms have a hemispherical "domed" shape. The center of the dome gets deflected to make contact with the seat and close the valve. Finite element analysis (FEA) and computer-aided modeling enables designers to systematically modify this traditional shape, along with the relative location of the seat and button, in the hunt for a configuration that provides the lowest possible diaphragm stress over the anticipated range of ALD process pressures. Hundreds of diaphragm valve designs may be created and evaluated in virtual space until an optimized configuration emerges with a predicted maximum diaphragm stress significantly below the fatigue limit of the diaphragm material.
If the target cycle life for an ALD valve is >25 million cycles, valve cycle life validation testing using standard procedures (e.g., 1 cycle per second) would consume more than a year (i.e., 32 million seconds). However, automated test equipment integrating electronic valve monitoring equipment and computer controls may be developed to enable high-speed cycle testing (e.g., 10 cycles per second). Any procedures used to accelerate valve cycle testing must be validated and shown to produce a failure population and failure modes that are consistent with that of standard testing and ultimately with the conditions anticipated in service.
In addition to cycle life, actuation speed is an important performance characteristic. Conventional wisdom has held that pneumatic systems are too slow and unreliable to provide the actuation speed and precision demanded by ALD processes. There are, however, inherent advantages to pneumatic actuation, including an installed base, smaller size, lower cost, safety, temperature capabilities, etc. A detailed assessment of pneumatic valve actuation demonstrates, to the surprise of many, that pneumatic systems can be used to achieve actuation response times of <;10 msec. The key to fast actuation is a combination of a quick, relatively high-flow solenoid pilot valve controlling the air supply to the pneumatic actuator; minimized air volume required to achieve actuation; and properly sized tubing to transport the air between the solenoid pilot and actuator. It turns out that the reputation of slow pneumatic response is an artifact of low-flow pilot valves, long tubing runs between pilot and actuator, and improperly sized tubing. Optimized pneumatic circuits, combined with an actuator design that minimizes the requisite air volume that needs to be displaced to actuate the valve, provide actuation response times that meet the needs of ALD applications, even those such as the LORA process that push the cyclic speed envelope. Figure 4 illustrates the valve actuation speed and repeatability that is readily achievable using pneumatics. Data substantiating the success of pneumatics is well received by ALD tool developers because of the familiarity, convenience and proven reliability of pneumatic systems, combined with the fact that alternatives to pneumatics — principally direct electric solenoid actuators — are expensive and have limited ability to operate in elevated-temperature environments common to ALD processes.
Historically, flow variation from one valve to the next has not been an issue. Mass flow controllers (MFCs) were employed to govern fluid flow. However, MFC response times are generally too slow to accommodate today's advanced ALD processes. In the absence of MFCs, diaphragm valves must provide consistent flow so that established processes can be readily duplicated on similar equipment and so process stability on individual systems remains when valves are replaced.
Another challenge associated with many ALD processes is temperature. ALD is a thermally activated and temperature-dependent process. Many of the precursors being tested and developed for optimized deposition of high-k dielectrics are solids or liquids at room temperature that need to be heated to achieve vapor pressures suitable for delivery to the process chamber. The fluidic components that carry these chemical precursors also need to be heated to avoid precipitation of the chemical back into its liquid or solid form. Too much heat, however, can damage the precursor. Some of the more attractive precursors require temperatures of M200°C to achieve useful vapor pressures.
Most fluid delivery systems on process equipment are not completely enclosed in a heated environment. The fluidic components (tubing, fittings, valves, etc.) are locally heated with cartridges, tape or thin-film heating elements to keep the process chemical at the desired temperature. An ALD valve needs to be compatible with high temperatures and also exhibit a uniform temperature distribution to eliminate problems associated with cold or hot spots in the flow stream. Achieving compatibility with elevated temperature requires a careful combination of material selection and valve design. Heating a valve body is complicated by the fact that it is connected to a pneumatic actuator. The pneumatic actuator pulls heat from the valve body and leads to large temperature gradients. Therefore, there must be some means of thermally isolating the valve body from the actuator.
ConclusionALD processes, especially those targeting high productivity via increased pulse frequency, place extreme performance demands on UHP diaphragm valves. Extending the capabilities of conventional UHP diaphragm valves to meet the requirements of ALD processes requires detailed knowledge of the ALD process, combined with cutting-edge design capabilities. The availability of valves tailored for ALD processes enhances the commercial viability and success of ALD.
| Author Information |
| Bill Glime, senior technical specialist, Swagelok new product development, is responsible for identifying and developing new product and process technologies that enhance the strength of the Swagelok product line. He has been with the company for seven years. He earned his Ph.D. in materials science from Case Western Reserve University in 1996. |
| Tom Seidel, CTO, joined Genus in 1995, and has been at the center of the development of ALD. At Bell Labs (1965-85), he contributed to advances in process characterization and device development: in microwave sources, ion implant and RTP technologies, and supervised a transfer technology for CMOS manufacture. Later, at Sematech, he held management positions for many development and strategic programs from 1988 to 1995. Seidel holds degrees from Stevens Institute of Technology (Ph.D.), Notre Dame (M.S.) and St. Joseph's (P.A. and B.S.) in physics. |
| References |
|



