Using mixed-fluid jet bombardment for advanced particle removal
03/01/2006
R&D studies evaluate the particle-removal performance and safe limits of single-wafer cleaning techniques based on either immersion megasonics or mixed-fluid jet bombardment. In leading-edge processes, results show that megasonic immersion is preferred for hydrophobic layers, while the mixed-jet approach provides more accurate control of mechanical agitation force for particle removal while protecting fragile structures on hydrophilic layers.
Steven Verhaverbeke, Roman Gouk, Erica Porras, Alexander Ko, Rick Endo, Applied Materials Inc., Sunnyvale, California
The number of cleaning steps required in wafer processing has increased with each device generation, reaching greater than 100 steps in some recently introduced process flows. The increase in the number of cleaning steps contributes to additional cycle time, cumulative silicon and oxide loss, and a greater potential to damage fragile structures. With continued transistor scaling in both horizontal and vertical dimensions, the challenges for cleaning are especially critical. While much attention has been focused on issues associated with the vertical scaling of the transistor above the silicon surface (e.g., the gate-oxide thickness), the vertical scaling inside the silicon surface (i.e., all junctions are getting shallower) is now also a concern in wafer cleaning steps. With shallower junctions, most of the implanted species are now close to the surface of the wafer.
Former cleaning processes etched part of the wafer surface and, in so doing, lifted a high percentage of particles off the surface. This method of particle removal cannot be employed in advanced IC processes because it would also remove most of the implanted species. The alternative to etching the surface out from under the particle is to use a mechanical agitation technique to knock the particles off the surface. However, the structures on the wafer surface are becoming more fragile because of horizontal dimension scaling, which results in a particle removal paradox: vertical scaling prohibits etching, while horizontal scaling prohibits mechanical agitation.
The most critical of all cleaning steps is the one that occurs after Ldd implant for lightly doped drains. This is true for two reasons. First, this cleaning step occurs when the smallest features on a chip are exposed (i.e., the gate is being etched and the spacers are not yet protecting the gate). Second, it is the step with the shallowest of all implanted species - the Ldd implant. Therefore, this step is often used as the barometer for advanced cleaning.
What is needed is an advanced particle removal technology that doesn’t rely on etching and can clean surfaces without damage. From our R&D work, we have shown that these requirements can be met by using a highly controlled mechanical force. However, the spread between the adhesion force of the patterns and the adhesion force of the particles is becoming narrower with each generation. Therefore, the force distribution of the mechanical agitation technique employed needs to be very narrow, repeatable, and well controlled.
A megasonic transducer mounted at the bottom of an immersion tank applied the mechanical agitation force traditionally used for particle removal. The typical problem with this approach is that it is virtually impossible to get a uniform, well-controlled agitation force over a batch of 50 wafers. However, a single-wafer approach allows the control and tailoring of a specific mechanical agitation force uniformly over the wafer.
Therefore, for mechanical particle cleaning with exposed fine patterns, single-wafer cleaning is mandatory. This requirement drives a major shift in clean processing from batch to single wafer, especially in the frontend-of-line (FEOL) steps, in which batch cleaning is still the most common approach. We examined a horizontal spin platform using both mixed-fluid jet agitation for removing particles as well as single-wafer immersion with various megasonic transducers to achieve accurate control over the particle removal force. The focus of this paper is on the mixed-fluid jet approach. The single-wafer immersion approach with multiple transducers was published earlier in Solid State Technology [1].
Mixed-fluid jet agitation
In the mixed-fluid jet approach, a liquid, which is usually DI water, is mixed with a gas - either N2 or clean dry air (CDA) - to form an aerosol that gets accelerated and bombards the surface of the wafer. The advantage of this technique is that the maximum velocity of the aerosol liquid droplets can be well controlled.
 Figure 1. Measured velocity and droplet diameter of the mixed fluid jet aerosol accelerated at 63 m/sec. |
The measured velocity and aerosol liquid diameter was measured with a phase Doppler particle analysis (PDPA) technique. As shown in Fig. 1, the maximum velocity of the liquid particles depends on the maximum gas velocity that accelerates them, which, in turn, depends only on the gas flow and the tube diameter. The gas flow is controlled by a mass flow controller and hence can be very accurate. The force on the wafer patterns depends almost solely on the velocity and is weakly dependent on the liquid aerosol diameter, since all of the liquid aerosol droplets are much larger than the patterns on the wafer. The pressure and flow patterns of a droplet hitting a solid wall can be calculated; the deformation and internal pressures inside such a droplet depend almost entirely on the impact velocity.
Experimentally, we confirmed that the droplet velocity depends on the N2 flow rate. We changed the N2 flow from 30 standard liters/min (slm) down to 20slm and recorded the velocity distribution of the droplets using the PDPA technique. From the distribution of droplet velocities, we can calculate the average velocity of the entire distribution. Both the average droplet velocity and the maximum velocity are directly proportional to the N2 flow rate over the range of 20-30slm.
The merit of this technique lies in the inherent control of maximum exerted force through controlling the gas acceleration. Even though there is a tail on the lower end of the impact force distribution, the higher end of the distribution is sharply controlled by the maximum gas velocity so that particle removal and pattern feature damage can be controlled (Fig. 2).
 Figure 2. Impact force distribution of mixed-fluid jet, and particle adhesion force distribution, together with polysilicon line strengths for 65nm and 90nm lines. |
null
 Figure 3. Gas velocity modeled from the nozzle exit to the wafer surface. |
Figure 3 shows the velocity modeling of the gas exiting the nozzle and hitting the wafer surface. In this nozzle, there is virtually no divergence of the accelerated aerosol stream, hence the height of the nozzle placement is not critical. Moreover, the gas velocity is in the turbulent regime and hence the velocity is uniform across the nozzle diameter. The particle removal efficiency (PRE) decreases only slightly from ~95% to ~88%, when the distance between the nozzle and the wafer is increased from 20mm all the way to 70mm.
Particle removal efficiencies of 85% on 0.1µm Si3N4 particles (aerosol deposited) were obtained with this technique without damaging 65nm poly-Si lines. Both particle removal and poly-Si line damage are dependent on gas velocity (Fig. 4). In general, both particle removal and damage of poly-Si lines increase with increasing gas velocity. However, as shown in Fig.4, below a gas velocity threshold of 44m/sec, there is no damage. It is clear from this figure that particle removal efficiencies (PRE) in excess of 80% and even 90% are possible without damage to fine structures.
null
Conclusion
For particle removal on fine patterns, batch cleaning can only use chemical undercut techniques since agitation cannot be controlled uniformly over a batch of wafers. In single-wafer cleaning systems, the removal force can be more uniformly controlled by either immersion megasonics or mixed-fluid jet bombardment. In a horizontal spin platform tool, a mixed-fluid jet technique offers very accurate control of the agitation force and has a narrow force distribution. Single-wafer systems provide major fine-line cleaning advantages over batch approaches, which up to now have remained in use for FEOL applications.
Acknowledgments
Thanks to Jim Papanu, Kent Child, and John Lee for their work and support with particle removal experiments, nozzle design, and discussion of the results.
Reference
- J.J. Rosato, M.R. Yalamanchili, “Using Multiple Transducers at Sub-65nm for Single-wafer Megasonics-based Cleaning,” Solid State Technology, October 2005, p. 50.
Steven Verhaverbeke is the CTO of the Cleans Product Group at Applied Materials Inc., 974 East Arques Ave., M/S 81307, Sunnyvale, CA 94085; ph 408/398-2587, fax 208/955-5580, e-mail [email protected].
Roman Gouk is a process engineer at Applied Materials’ Wet Clean Group, focusing on process development for end-user applications.
Erica Porras is working in the New Products Group at Applied Materials Inc.
Alexander Ko is a member of the technical staff in the Cleans Product Group at Applied Materials.
Rick Endo is an engineering manager at Applied Materials.