Process and environmental benefits of HF-ozone cleaning chemistry
07/01/2001
Eric Bergman, Sébastien Lagrange, Semitool Inc., Kalispell, Montana
overview
HF-ozone cleaning chemistry can perform the same functions as a conventional four-chem clean — organic removal, particle removal, chemical-oxide strip and regeneration, and metal contamination removal. This process chemistry also reduces cycle time and requires significantly less complex hardware. Properly configured HF-ozone chemistry saves cleanroom floor space and, because it uses low volumes of chemicals, it further reduces the environmental impact of semiconductor manufacturing. HF-ozone cleaning chemistry may become a method of choice for cleaning applications in single-wafer processing.
 Figure 1. Residual carbon, oxygen, and silicon after stripping photoresist-coated silicon wafers using four different strip chemistries. |
Most cleaning chemistry used in wafer processing is based on the "four-chem clean": (1) removal of organic contamination using a sulfuric acid and hydrogen peroxide mixture (SPM or "piranha"), sulfuric acid and ozone mixture (SOM) at 100-130°C, or deionized (DI) water and ozone mixture at subambient temperature with typical wafer exposure times of 3-10 min; (2) removal of sacrificial silicon dioxide or native oxide using a 50:1 or 100:1 dilution of 49% hydrofluoric acid (HF) at ambient temperature for 1/2-1 min; (3) removal of particle contamination and reoxidation of the silicon surface using a DI water, ammonia hydroxide, and hydrogen peroxide water mixture (APM or "SC1" or "RCA1"), commonly a 20:4:1 dilution, with wafer exposure times from 3-10 min and, frequently, with megasonic agitation and temperatures from 65-80°C.; and (4) removal of metallic contaminants using a DI water, hydrochloric acid, and hydrogen peroxide mixture (HPM or "SC2" or "RCA2"), commonly a 5:1:1 dilution, with a 90-sec wafer exposure at 65-80°C.
In some variations of applications, silicon dioxide chemically grown during the APM step may be removed using an HF immersion or HF vapor process. However, debate continues on the need for a hydrophobic surface prior to growing thin gate oxides [1, 2].
Because rinsing is required between each of these chemical steps, the total process sequence is often in the range of 20-40 min. Equipment setups for these chemistries are complex and chemicals and waste disposal are costly. In addition, the industry continues to express interest in performing critical cleaning processes using single-wafer processing. However, at 20 min/wafer, this becomes impractical. Consequently, there is a need for a simplified chemical cleaning process that will perform all four-chem-clean functions.
Boundary layer control plus HF
We have found some benefits with a process — HydrOzone [3, 4] — that creates a controlled boundary layer on a wafer surface by combining a controlled spray, a steam condensate, wafer rotation, and flow control. Injecting ozone into the process chamber surrounds the wafer with an ozone-rich environment. The ozone then readily diffuses through the controlled boundary layer, resulting in a surface ozone concentration and subsequent oxidation rates significantly higher than what can be achieved using ozone dissolved in water. In addition, this technique overcomes ozone solubility limitations imposed by temperature; like all gasses, the solubility of ozone in an aqueous solution is inversely proportional to temperature. By circumventing this limitation, we have achieved photoresist strip rates >1µm/min, although rates in the range of 8000Å/min are more common.
While this process is highly effective at oxidizing organic materials, including photoresist, it fails to address other criteria of a comprehensive chemical clean. These include metallic ion contamination, particulate contamination, and removal and regeneration of a passivating oxide film. However, when we added HF to the HydrOzone process — creating a FluorOzone process — we found that we could effectively overcome deficiencies.
In developing the chemistry for this new cleaning concept, we had to consider a number of issues. For example, the targeted rate of thermal oxide (TOX) etch should be able to remove the desired amount of oxide in a reasonable time. With the four-chem clean, the oxide thickness removal target is usually 20-30Å. Therefore, the targeted process time for HF chemistry delivery should be 2-5 min or an etch rate of between 4 and 15Å/min. Based on this, we defined an HF concentration of ~0.1% (a 500:1 dilution of 49% by weight HF) as standard. This chemistry has an etch rate on TOX of ~6Å/min at 95°C, so the targeted process time ranges from 3-5 min of chemical delivery.
Our preliminary testing indicated a correlation between HF temperature and particle removal efficiency, particularly for silicon dioxide particles. Highest efficiencies were attained at high temperatures, so we defined 95°C as standard. High temperature also promotes chemical reactivity with metallic ions.
We found that our new chemical cleaning process was critically affected by etch uniformity. We determined that rotational speeds from 25-50rpm typically delivered optimum etch uniformity. A series of photoresist strip rate tests showed an increase in strip rate with respect to RPM, but a decrease in rate with increasing flow rate. So while the lower rotational speed creates a thicker boundary layer, it is still sufficiently thin to permit ozone to diffuse and react with silicon or organic materials at a waferÅs surface.
Due to dynamic conditions within the process chamber, it has thus far not been possible to precisely define liquid boundary layer thickness on spinning wafers. Metal removal efficiency is actually quite high with just diluted HF, particularly at elevated temperature, compared to a typical HPM clean. However, we found that we could attain optimal metal contamination removal by adding a small amount of HCl to the cleaning solution, simultaneous with HF; we found that a 500:1 dilution of 30% HCl was effective.
With this chemistry, we maintain ozone concentration at a high level to drive diffusion through the boundary layer, oxidize organic materials, and oxidize the silicon surface to ensure the wafer remains hydrophilic at all times. Even when spraying HF solutions with concentrations high enough to achieve a TOX etch rate >120Å/min, we found that a hydrophilic surface was still maintained on bare silicon wafers, illustrating the ability of ozone to diffuse through the boundary layer.
We have used three different techniques to deliver ozone and HF and HCL process chemistry. In the first, we used a syringe pump injection system to deliver HF and HCl into a fresh, temperature-controlled DI water stream just prior to spraying onto wafers. Ozone can also be injected in this line, or it can be brought in through a separate manifold. Chemical concentration was controlled by controlling injection rate and DI water flow.
In the second technique, we used a recirculating tank of diluted HF or HCl that fed a spray into a chamber and reclaimed the fluid. While this method was relatively simple, it ran a slight risk of accumulating contaminants in the tank. This type of technique has been used with considerable success in wafer processing for several years, however.
Using a Semitool Capsule process system, we were able to recirculate the contents of a chemical tank through an ozone infusion system (i.e., a semipermeable gas membrane) to saturate the liquid with ozone. In this case, the available ozone concentration may not be as high, and temperature may be reduced to promote ozone solubility. Because this is a single-wafer process, a higher TOX etch rate may also be used to shorten process times. The batch spray process is strictly a hydrophilic process (i.e., ozone regenerates a silicon dioxide surface as quickly as HF etches it away, thereby maintaining oxide thickness). The Capsule process, however, can be configured to generate a hydrophobic surface that is then dried via an IPA vapor-generated surface tension gradient under centrifugal force.
Process results
To evaluate our new chemistry for cleaning, we first compared residual surface-carbon contamination following photoresist coat and strip using four different methods (Fig. 1). We found that our HF-ozone process was comparable to SPM, which is the industryÅs standard organic cleaning chemistry. While it appears that dry-ozone and oxygen-plasma ashing have even lower carbon levels, we feel that this is most likely an artifact of higher oxygen content at the surface due to bond formation in a more energetic environment. The sum of carbon, oxygen, and silicon adds up to 100%, so an elevation in any one category must necessarily suppress other categories.
Next, we compared metal-cleaning capabilities of our chemistry to the standard HPM clean. Wafers were intentionally contaminated with iron, nickel, and copper to ~1 x 1014atoms/cm2. While HPM was able to reduce contamination down to ~5 x 1010atoms/cm2, our HF ozone process reduced these even further, down to the detection level of vapor phase decomposition ion-coupled plasma mass spectroscopy; our data showed that post-clean metals were <1 x 1010atoms/cm2 for all metals except for iron after the SC2 clean and HydroOzone (HCl) chemistries.
While a modified version of the conventional four-chem clean, which included the conventional chemistries but with additional chemical cleaning steps, was capable of reducing silica slurry particle counts from ~40,000 0.15µm particles down to 2000-3000 during the course of a 30-min process, our HF-ozone chemistry achieved an order of magnitude improvement in just a 5-min process. When we followed our process with a conventional APM chemistry, also delivered as a spray process step, final particle counts were reduced even further to an average of 56 (Fig. 2).
We found that the SC1 clean alone was capable of achieving particle removal efficiencies on dried silica slurry of ~25%. The four-chem-clean performance was ~95%, while the HF-ozone process improved removal efficiency to 99.5%. Combining HF-ozone with APM resulted in a 99.8% removal efficiency; it also reduced process time by >50%, water consumption by 75%, and chemical usage by >80%.
 Figure 3. Three independent tests to monitor surface roughness of <100> silicon wafers measured with AFM after 30-min HF-ozone exposure at 95°C. |
Interestingly, the results were somewhat different for silicon nitride particles. While numerous tests have shown that these can be removed with fluorinated chemistries, the results with our HF-ozone process were somewhat erratic. Removal efficiency varied from 25-75%, with wafers prepared in accordance with Sematech Technology Transfer No. 94052368A-XFR. However, when the ozone supply was intermittently cycled on and off during the course of the cleaning step, cleaning efficiency increased to >99%. While this remains to be studied further, it appears that ozone concentration reduction in the process chamber during off cycles permits temporary thinning of the chemical oxide on silicon, which is advantageous in detaching silicon nitride particles. In any case, it was demonstrated that HF-ozone particle removal efficiency was superior to APM.
Because HF will etch exposed silicon dioxide surfaces, etch uniformity is critically important for viable processing. In our studies, as noted above, the anticipated TOX etch target was 20-30Å. It is difficult to obtain an accurate view of etch uniformity when removing such a small amount of oxide, since a one-atomic-layer measurement variation will contribute to a perceived process nonuniformity of up to 10%. Consequently, we used an etch target of 250Å to test etch uniformity; our data showed a 251Å (1.19% 3s) etch uniformity using dilute HF and ozone at 95°C.
Additional etch uniformity studies targeting 30-40Å of silicon dioxide removal have shown etch uniformity and repeatability to be in the range of one atomic layer. As stated previously, the process can be run in either a recirculated bath mode or using fresh water with HF and HCl injected into the delivery stream. Because the HF bath is operated at temperatures up to 95°C, we were concerned about bath stability in the form of HF losses through evaporation over time. However, we learned from studies that stability can be exceptional throughout 96 hr. Because the vapor pressure of HF over the solution is a function of the mole fraction of HF in solution, and because such an exceptionally diluted solution is being used, the change in bath concentration due to evaporative losses is slight. In fact, it was observed that etch rate actually increases slightly over time, indicating that the rate of water loss is proportionally greater than the loss of HF.
Because the HF will etch exposed oxides, and because the presence of ozone will oxidize exposed silicon, an HF-ozone chemistry effectively becomes a silicon etchant. This etch rate is controlled by HF concentration as long as ozone concentration is maintained at a level that will generate silicon dioxide faster than HF will etch it. If ozone concentration is reduced and HF concentration elevated, oxide growth rate essentially becomes the limiting step. Since the objective in a spray process is to maintain a hydrophilic surface, low HF concentrations and high ozone concentrations are typically used. So, we performed tests to ensure that the surfaces of wafers were not significantly roughened by our process (Fig. 3).
In tests depicted in Fig. 3, we ran our HF-ozone process for 30 min in an effort to aggravate silicon surface roughness. Neither the RMS surface roughness nor the max.-to-min. surface roughness appears to have been adversely affected. However, in tests using control wafers with an RMS surface roughness of <1Å, the surface roughness was seen to increase, with HF-ozone-processed samples, by ~0.5Å. Surface roughness was approximately the same for both a 5-min and 30-min process. At this time, we do not know if this will lead to an adverse impact on device performance. However, tests have been performed that indicate minor differences in chemical oxide quality and growth technique as seen immediately after cleaning. This may have no discernable impact following the growth of gate oxides as thin as 16Å in an RTP system [5].
We also studied polysilicon-to-thermal-oxide etch selectivity. Results indicated a 1:1 selectivity even when polysilicon was doped to ~1 x 1015atoms/cm2. Even though surface roughness does not appear to increase, it must be recognized that a finite amount of silicon and oxide loss will be incurred, so device performance will have to be evaluated.
As noted previously, our HF-ozone process produces a hydrophilic silicon surface. In the event that a hydrophobic surface is required, it must be generated in a subsequent processing step, most typically via HF immersion or vapor to maintain the particle integrity of a cleaned surface. While spin-rinse-dry methods produce excellent results on hydrophilic surfaces, a surface tension gradient dry is preferred for hydrophobic surfaces.
Footprint, environmental relief
A four-chem clean packaged in a typical immersion wet deck can be in excess of 20 ft long. By comparison, an HF-ozone process can be packaged in a spray platform that is only 2 ft wide.
If the HF-ozone process were configured for all fresh chemicals, a 50-wafer load processed in the dry-to-dry mode would require approximately 56 liters of water and 48 ml of 49% HF. In a recirculated system, the chemical consumption would be 0.2 ml of HF per 50-wafer run and only 32 liters of water. Contrast this with a four-chem-clean process requiring a 40-liter piranha tank with an estimated life of 12 hr for a usage of 555 ml
un. This does not include a peroxide addition "spike," 100:1 HF with an estimated bath life of 24 hr for a usage of 277 ml of 100:1 HF or 2.7 ml of 49% HF
un. In addition, a four-hr bath life for APM and HPM would result in the consumption of 66 ml of NH4OH
un, 504 ml of H2O2, and 238 ml of HCl. Total chemical consumption would then be 1365 ml
un. Assuming a 5-min rinse after piranha, HF, APM, and HPM at a minimal volume of 20 liters/min, and the four-chem-clean water consumption is 400 liters/50-wafer run.
Conclusion
A HF-ozone cleaning chemistry performs the same functions as a conventional four-chem clean: organic removal, particle removal, the strip and regeneration of a fresh chemical oxide film, and the removal of metal contamination. In addition, it reduces cycle time and requires less complex hardware. Properly configured, this process saves cleanroom floor space and reduces negative environmental impact. Also, since single-wafer cleaning cannot be done efficiently with conventional chemistries, HF-ozone cleaning chemistry may become the method of choice for cleaning applications in single-wafer processing.
Acknowledgments
The authors thank Martine Claes and Erika Rhör of IMEC, Leuven, Belgium, for their assistance in characterization and consultation. FluorOzone and HydrOzone are trademarks of Semitool Inc.
References
- J. Jeon et al., "Characterization of HF-Last Cleaned Si for Gate Oxides," Cleaning Technology in Semiconductor Device Manufacturing V, The Electrochemical Society, Proceedings Vol. 97-35, p. 195.
- A. Bayoumi et al., "Investigation of the Need for Alternative Cleaning Chemistries for 30Å Gate Oxides," Science and Technology of Semiconductor Surface Preparation, MRS Symp. Proc., Vol. 477, p. 247, 1997.
- S. De Gendt et al., "Evaluation of Ozonated Water Spray for Resist Cleaning Applications," Cleaning Technology in Semiconductor Device Manufacturing V, The Electrochemical Society, Proceedings Vol. 99-36, p. 391.
- E. J. Bergman, M. Melli, M. Magrin, "Photoresist Strip Process Using Ozone Diffusion Through a Controlled Aqueous Boundary Layer," Cleaning Technology in Semiconductor Device Manufacturing V, The Electrochemical Society, Proceedings Vol. 99-36, p. 399.
- J. Eng, Jr., et al., "The Evolution of Chemical Oxides into Ultrathin Gate Oxides: Spectroscopic and Electrical Characterization," Fifth Intl. Symp. on Ultra Clean Proc. of Silicon Surfaces UCPSS 2000 Abstract Book, p. 106.
Eric Bergman received his BS in chemical engineering from Brigham Young University. Bergman is a staff engineer at Semitool Inc., 655 W. Reserve Dr., Kalispell, MT 59901; ph 406/752-2107, fax 406/752-5522, e-mail [email protected].
Sébastien Lagrange received his masters in chemistry from Ecole Nationale de Chimie de Toulouse in France. Lagrange is on assignment, from Semitool, at IMEC in Leuven, Belgium, where he manages joint development projects in cleaning technology and copper metallization.