Dilute RCA Cleaning Chemistries

The development of dilute cleaning chemistries has peaked industry interest. As market competitiveness increases, chemical consumption and associated costs become increasingly important. Dilute ammonium hydroxide (NH₄OH)/hydrogen peroxide chemistries (SC-1) have the advantage of reduced oxide loss while preserving excellent particle removal performance. Similarly, many studies have shown that highly dilute hydrochloric acid/hydrogen peroxide (HCl)/(H₂O₂) mixtures (SC-2) are effective at controlling metallic contamination.^1,2,3. This article will demonstrate the effectiveness of this cleaning technique in a production environment.

The RCA clean sequence developed by Werner Kern⁴ in the 1960s still is used widely in semiconductor manufacturing as a critical clean for the removal of organic, metallic and particulate contamination on wafer surfaces prior to oxide growth operations. Only SC-1 and SC-2 portions of the clean sequence will be discussed.

High pH SC-1 is an effective particulate removal chemistry, aided by the high negative zeta potential of both silicon and oxide in this pH range. SC-2 is effective at removing metallic contamination with a pH low enough to ensure good metal oxide solubility and with the chloride ion acting as a complexing agent.

Comparisons were made between traditional SC-1 chemical ratios (1:1:5, NH₄OH:H₂O₂:H₂O) and dilute chemistries (1:4:20) at 60°C. The dilute chemistries used megasonics. Similarly, comparisons were made between traditional chemical ratios for SC-2 (1:1:6, HCl:H₂O₂:H₂O) at 85°C and dilute chemistries (1:1:20) at 60°C. Defect densities, oxide loss and capacitor breakdown were assessed.

Defect Density

Evaluation of particle removal efficiency of traditional SC-1/SC-2 processes to dilute SC-1 with megasonics (DM SC-1) and dilute, lower-temperature SC-2 (DLT SC-1) processes produced encouraging results. Testing was done on both bare silicon and thermal oxide monitor wafers using a Tencor 6400 with a minimum equivalent particle size of 0.16 µm. During experimental evaluation, both bare silicon and thermal oxide showed a trend toward improved particle removal with the DM SC-1 and DLT SC-2 chemistries. Results from production pre-gate oxidation tool qualification testing are shown in Figure 1. A comparison of the average number of particles added to 200 mm bare-Si wafers indicate improved defect density with dilute chemistries. Reduction in poly silicon defect densities further support this claim.

Fig. 1. Comparisons of production pre-gate oxidation cleans indicate lower defect densities when using dilute chemistries.

Another significant measure of improved particle performance is the poly silicon defect density. A combined DM SC-1 and DLT SC-2 pre-gate oxidation clean typically is followed immediately by gate oxidation and poly silicon deposition. In this case, defects present on the wafer surface as a result of pre-gate oxidation clean were measured by the Tencor tool. A 2X reduction in poly silicon defects was demonstrated.

SC-2 is not effective at surface particle removal; however it may impact particle redeposition. Both silicon and oxide have relatively low zeta potential in acidic solutions; therefore SC-2 at either 1:1:6 or 1:1:20 is not effective at preventing redeposition. Traditional SC-2 is a much more effervescent solution than dilute SC-2. Effervescence may contribute to an increased particle deposition rate. Therefore, the lowered effervescent effect of dilute SC-2 appears to result in a lower particle deposition.

Reductions in redeposition of removed particles at the SC-1 level can be achieved with dilute chemistries also. Because of the significant dilution of SC-1, the pH remains high enough to maintain the high negative zeta potential of silicon and oxide while reducing the ionic strength of the solution.

A slight undercutting of surface particles is believed to be the primary mechanism for removing surface particles in SC-1. Since dilute chemistries have been shown to reduce the oxide etch rate, the addition of megasonics may reduce the amount of undercutting required to achieve equivalent particle detachment. The implication is that megasonic energy can improve particle removal efficiency by increasing effectiveness of the surface etch.

Oxide Loss

Fig. 2. The reduction in thermal oxide loss is attributed to the reduced ammonium hydroxide concentration in dilute SC-1.

Traditional and DM SC-1 baths were compared for their contributions to thermal oxide loss. Figure 2 illustrates a 10% reduction in oxide etch rate in dilute megasonic SC-1 as compared to traditional SC-1. Reduction in etch rate is due to a lower concentration of ammonium hydroxide in dilute SC-1. However, SC-2 chemistries have little effect on oxide loss and were characterized for reference only. The dilute megasonic SC-1 process for this test preceded SC-2 oxide removal testing.

Atomic force measurements were performed on thermal oxide surfaces processed in both traditional and dilute megasonic SC-1 to determine the effect on surface roughness. No significant difference was noted between the two approaches. However, variations in the surface roughness of the silicon substrates were larger than expected contributions from the clean.

Gate Oxide Integrity

One concern when considering any changes to the pre-furnace wafer cleaning process in a CMOS fab is the impact on gate oxide integrity. Changes in the ability of the cleaning process to remove particulate contamination or metallics potentially could degrade oxide quality. It has been reported that the diluted SC-2 concentration used here may not be as effective as the conventional solution at removing elevated concentrations of Al or Ni ⁵.

Fig. 3. Breakdown distributions from (top) lots using dilute chemistries prior to gate oxidation are comparable to conventional cleans and (bottom) indicate similar results in product runs. Significant reduction in spiking volumes can be achieved in dilute chemistries.

To evaluate the effect of dilute-cleaning on oxide integrity, a split lot containing MOS capacitors of various geometries was run for the wafer clean immediately prior to gate oxidation. The lot was split between traditional SC-1 and SC-2 cleans and DM SC-1 and DLT SC-2 cleans. The wafer lot then was recombined for gate oxidation and the remainder of the fabrication process. Capacitors of ~10,000 µm² were tested to evaluate intrinsic oxide quality, and larger area capacitors, ~300,000 µm², were tested to assess changes in the defectivity of the oxide. A standard voltage ramp, destructive breakdown test was used to quantify the results. Breakdown data from a limited production run were also collected from a larger sampling.

Figure 3 shows cumulative probability plots for the oxide breakdown field for both large and small area capacitors of the test lot and of limited production runs. The data indicated that the diluted cleaning process maintained good intrinsic oxide quality and produced low defectivity in the oxide. The data from the production run demonstrated that dilute cleans performed as well as conventional cleans in a manufacturing environment.

Bath Control

Control of SC-1 and SC-2 bath chemical ratios is crucial to maintaining optimum performance of the clean operation. Tight control of the peroxide concentration in SC-1 is important for management of oxide loss, minimization of chemically induced surface roughening and silicon attack prevention. Accurate levels of peroxide in the SC-2 bath are critical to the oxidation of noble metals such as copper.

In this study, the TEL UW8000 wet bench was used to control initial bath concentrations by filling the process bath from premeasured canisters of preheated ultra-pure water, reagent and peroxide. The premeasured volume of each canister was adjusted and verified by assay of the poured up bath. Volumes of reagents and peroxide were periodically meter/spiked. The bath level was monitored and a full level maintained with ultra-pure water.

Fig. 4. Dilute, low-temperature SC-2 exhibits little peroxide decomposition, while significant decomposition is seen at traditional concentrations.

Fig. 5. SC-2 bath concentration can be controlled by periodic spiking of HCl and hydrogen peroxide.

Level loss in the bath is caused primarily by drag-out when wafers are removed from the baths. The volume of water is not defined by application but based on the amount of water needed to reach full level. The level adjustment with water and the addition of water from hydrogen peroxide decomposition contribute to dilution of SC-1 and SC-2 baths. The dilution of the bath is more significant at traditional SC-1 and SC-2 concentrations than at the diluted concentrations evaluated for this process.⁵

The effects of bath level maintenance and peroxide decomposition on percent by weight of peroxide in an SC-2 bath are shown in Figure 4. The hydrogen peroxide concentration in the dilute SC-2 bath increased at 60 min, because the spiking frequency was set at that time frame. It should be noted that the peroxide spiking volumes of ~1% of the total bath volume were insufficient in preventing peroxide depletion in the traditional SC-2 bath. In contrast, peroxide decomposition essentially stops with dilute concentrations and at lower SC-2 temperatures, even with no peroxide spiking.

Additional experimental work is required to fine-tune the reagent and peroxide spiking volumes and frequencies for further improvement in bath composition control. Figure 5 shows the results of the first iteration of reagent and peroxide spiking parameters on the SC-2 bath. The bath composition was verified by drawing samples from the bath for assay.

Chemical Usage

SC-1 and SC-2 baths can reduce chemical usage significantly. This means lower volumes of reagent and hydrogen peroxide required for mixing DM SC-1 and DLT SC-2 baths than for traditional mixtures. Because chemical concentration is reduced, and hydrogen peroxide decomposition is no longer a limiting factor, significant reduction in spiking volumes can be achieved in dilute chemistries. The dilute process discussed has a bath changeout frequency of 4 hrs for SC-1 and 6 hrs for SC-2, resulting in a 40% reduction in chemical costs for SC-1 and 90% for SC-2. Increased stability in This calculation assumes full production at continuous 24 hr operation of UW8000 wet benches. Further enhancements and process optimizations may yield longer chemical lifetimes.

Concluding Statements

A diluted RCA clean process has been developed and implemented for pre-gate oxidation cleans in a proven manufacturing environment. This advancement in wafer cleaning technology has resulted in decreased defect densities, improved oxide loss control in SC-1 and significant reduction in chemical usage with no adverse effect on gate oxide integrity.

References

1. T. Q. Hurd, SEMI 1995 - Cleaning Technology for the Sub mm Era, p. 69, 1995.
2. T. Q. Hurd, P. W. Mertens, H. F. Schmidt, D. Ditter, L. H. Hall, M. Meuris, M. M. Heyns, 1994 Proceedings of the Institute of Environmental Sciences, p. 143, 1994.
3. T. Q. Hurd, P. W. Mertens, L. H. Hall, M. M. Heyns, UCPSS 1994 Proceedings, p. 435, 1994.
4. W. A. Kern and D. A. Poutinen, RCA Rev., p. 187, 1970.
5. 5. D. J. Riley, J. S. Glick, V. Parks, G. Matamis, The Impact of Temperature and Concentration on SC-2 Cost and Performance in a Production Environment, MRS Symposium Proceedings, Vol. 477, p. 519, 1997, editors M. Hirose, S. Raghavan, S. Verhaverbeki.

Acknowledgements

The authors wish to thank Advanced Micro Devices FAB25 staff and management for their support of this work.