Wafer Drying in Wet Processing: The Challenge of the Future

Semiconductor manufacturing has faced tremendous changes in wafer sizes and chip features over the past 15 years, and will continue on this track for years to come. Wafer sizes have increased from 100 to 300 mm for the mass production of processors and memory chips, and the level of integration has risen along with the shrinking of feature sizes.
 
This dynamic process has also included a lot of trial and error in the development of various kinds of process equipment. Nevertheless, despite all the discussions and questions, wet processing is still state-of-the-art and will remain so for at least the next decade. Mostly, it is used for cleaning purposes and surface preparation, as well as etching, whether single-wafer or batch treatment. Decreasing feature sizes and more complex structures demand even more process steps and higher requirements, especially for wet process performance.

Within the long line of process steps, drying is the most important "wet process" step. The drying step, which is done at the end of almost every wet process sequence, contributes much to the wet process performance, perhaps even defining its results.

In addition to an increasing number of 300 mm wafers being processed, vertical integration and shrinking critical dimensions (CDs) create new challenges for drying. Some production is already being done at the 90 nm node, and major chipmakers are involved in R&D for the 65 nm node. Accordingly, aspect ratios of 50-60:1 have become standard for deep trenches, and advanced applications are already at ratios of up to 100:1. These complex wafer geometries and structures have led to very mechanically fragile structures, redefining some of the major wafer drying performance requirements:

• Smaller particles sizes come into focus. The technology should work on particle levels of 0.1 µm diameter, with less than 20 adders on hydrophilic and hydrophobic surfaces.
• Mechanical stress introduced to the wafers during any process step must be reduced or completely avoided. This more or less defines the end of spin dryers for advanced applications.
• High-aspect-ratio structures, new materials and a mix of hydrophobic and hydrophilic surface areas require new process settings and a high flexibility to adapt the parameters to each particular need.
• Additional demands from customers call for combined drying and wet etching in one tank for complex applications.

There are various drying technologies today that are used throughout the industry. Each has its own pros and cons, but not all of them have the inherent potential to meet the demands of future technologies. In fact, all of them are reaching their limits, whether in terms of mechanical stress, achievable particle level or process performance, as well as the usable process window.

Spin dryers are still widely used in chip production. Because of the mechanical stress introduced to the wafers, it is clear that this technology cannot be used for the advanced applications of future chip production. IPA-vapor dryers are also commonly used today. But heated isopropyl alcohol (IPA) represents a major safety risk, which calls for a lot of expensive counter messages like CO₂-fire suppression systems. Not all accept this. A high chemical consumption, special efforts for IPA reprocessing, and total organic carbon (TOC) within the deionized water increases the cost of ownership of this technology. In addition, for some niche applications, special drying techniques have been developed (i.e., the HF/ozone dryer), but these are not applicable or accepted for a wide range of chip or material types.

The only technology that already has inherent potential for the future is the drying technology based on the Marangoni effect. This drying technology has been well established over the past dozen years and accepted industrywide.^1-8 But the commonly used technical solutions for this drying technology have also reached their limits. The low amount of IPA currently available within the process chamber and the inability to change it on a recipe base requires changes and improvements, especially for larger wafer sizes and smaller CDs. Furthermore, the movement of the boundary layer must be improved to overcome heavy and complex mechanisms.

We targeted Marangoni-style drying for revision because of its potential for further improvements. The most important step seems to be a new technique to introduce the alcohol into the process chamber. A major breakthrough was achieved with the creation of aerosol by ultrasonic oscillation. Ultrasonic nozzles offer several potential benefits such as a soft low-velocity spray, micro-flow and extensive spray shape capabilities, and total freedom from clogging. The aerosol is created by an ultrasonic oscillator that operates on a specific resonant frequency, primarily defined by the mechanical length of the nozzle. With a frequency range of 25-120 kHz, the nozzle length varies from 200 down to 40 mm. To produce atomization, a standing sinusoidal longitudinal wave is required for a sustained vibration. For the droplet size, the ultrasonic frequency is the predominant factor besides the surface tension and the density of the liquid used. Finally, not only one particular droplet size is created; the size follows a distribution function where you can find a number of droplets with greater and smaller diameters. Currently used oscillators typically emit droplets at an average diameter of 20 µm (Fig. 1 ).⁹

1. Droplet distribution (logarithmic) for the ultrasonic atomizer US1 from Lechler Comp. (water @ 0.02 lpm feed flow rate and 100 kHz).10

2. Cross-section of an ultrasonic nozzle.9

3. Visualized IPA aerosol introduction into the open process chamber, injected from an ultrasonic atomizer mounted within the process chamber lid (flow pattern for test purposes only). (Source: AP&S)

The oscillation is driven by a piezoelectric transducer, which is mounted on a very high-strength titanium alloy (Ti-6Al-4V). In the center of the nozzle a feed line leads the liquid to the bottom or free end of the oscillating part (Fig. 2). Depending on its shape, a particular fog geometry (i.e., conical, flat, narrow-jet-like) can be created to introduce the aerosol into the process chamber (Fig. 3 ). The overall efficiency of the atomization is an optimization of several factors such as input power, nozzle type, liquid characteristics and the feed flow rate. A minimum or critical power level is required to start the atomization — to expel droplets. If the power level exceeds the optimum, the material is literally ripped apart, causing large chunks of material instead of a soft spray of fine droplets.⁹

These methods gain a couple of new process parameters, which now can be used for the recipe. The major one is the amount of alcohol — the feed flow rate to the nozzle. Currently, IPA is used for the alcohol. Alcohol, especially IPA, as a pure single-component liquid with low viscosity and surface tension, is very suitable for use with an ultrasonic nozzle. Today, this technology has yet to find a process limitation. Besides the amount of alcohol itself, defined by the feed flow rate to the ultrasonic oscillator, there is also an option in the future to optimize ultrasonic frequency and amplitude of oscillation for best droplet sizes for existing or other alcohols.

Already, the variation of the alcohol amount opens up a huge process window, now available for the process settings (Fig. 4). Feed rates to the oscillator are possible from <1 mL/min up to 20 mL/min for one hardware configuration. This leads to an improved drying process control by the tool controller. Now the amount of alcohol used to dry the wafers, as well as the thickness of the alcohol layer on top of the DI water surface, which is moved across the wafer surface becomes a variable process parameter. Even this increased amount of alcohol is still far less than what is considered major chemical consumption. During test, it was found that today's standard applications on 200 mm wafers require only a very low alcohol use (5-10 mL/run depending on the CD and the number of wafers to be dried). Figure 5 shows a dried watermark-free hydrophobic wafer surface; a few droplets could be found only at the area of the edge exclusion.

4. Comparison of the typical process window of the common surface-tension-gradient (STG) drying technology vs. the new AeroSonic differ-ential-surface-tension (DST) drying technology.

5. Microscopic photographs of test wafers (110 nm technology) with Si3N4 layer after HF-last process with aerosonic dry. (Source: Infineon)

The second major process parameter for this drying technology is the so-called drain speed. This is the speed at which the boundary layer (a very thin layer of alcohol on top of the DI water surface) moves across the wafer or substrate surface. A drain speed window of 0.3-2.5 mm/sec is available, with up to 3.5 mm/sec possible. Today's typical setting is ~1-1.5 mm/sec. With the availability of different drain speeds over the bath cross-section, different wafer surface areas can be treated separately or water columns on top of the wafers can be drained faster to gain time.

First, intensive tests were carried out within the company's process lab, followed by an evaluation on the production side. Particle counts of <10 ppw @ 0.12 µm and <20 ppw @ 0.10 µm could be achieved (50-wafer PEC/200 mm). During these tests, users especially liked the options for the process control and flexibility. This also makes the technology attractive for extensive use in R&D applications. Furthermore, rinse and drying within a standard process bath further opens the possibility for integrated chemical steps for surface preparation before drying or the addition of surfactants to reduce surface tension for small structures. This special one-bath immersion processing with integrated drying step leads to potential reductions of tool sizes. A typically used surface preparation step is the HF-last process right before drying, where any exposure of the hydrophobic wafer surface to air should be avoided.

With the very low alcohol utilization during the process, the new technology avoids safety risks; the IPA concentration stays far below its flame point. The processes run under ambient process conditions within an inert atmosphere within the process chamber by N₂.

Tailored drying solutions are attractive for silicon or IC manufacturers. Adapted processes can be set for a broad range of substrate materials and sizes, just by changing the different recipe parameters within the tool software (Fig. 6 ). Materials such as silicon, SiO₂, quartz and GaAs come into focus, as well as SiC or other new materials. It is also interesting for MEMS, GaAs, bumping or other substrate drying (FPD substrates).

6. Average performance requirements for different industrial applications, where a drying technology is needed.