M. Riva, M. Pittroff, T. Schwarze, J. Oshinowo, Solvay Fluor GmbH, Hannover, Germany; R. Wieland, Fraunhofer Institute for Reliability and Microintegration (IZM), Munich, Germany
A new F2 gas mixture was evaluated as a substitute for conventional cleaning gases such as, NF3, C2F6, and CF4, in a CVD chamber. The new mixture was compatible with the equipment used and improvement was seen in the etch rate as well as a reduction in the amount of gas needed to complete the clean.
F2 gas mixtures offer ideal properties as chamber cleaning gas: low dissociation energy and high reactivity, which leads to superior efficiency and ease of abatement. In this work, a new F2 gas mixture with a combination ratio of 10% Ar, 20% F2 and 70% N2 was used to obtain a maximum of 20% fluorine in inert gases. This mixture has been evaluated as a candidate to replace conventional cleaning gases such as NF3, C2F6, and CF4. The tested Ar/N2/F2 mixture showed improvements in both parameters, cleaning up to 27% faster, and requiring 96% less gas than when using NF3. The etch–rate performance of the Ar/N2/F2 gas mixture was combined with etch nonuniformity values of ±3% (1σ) on SiO2, and ±8% (1σ) on Si3N4, respectively. Additionally, amorphous silicon (a–Si) was etched completely and uniformly. The particle performance data showed an average of 14 particle adders (0.25μm), indicating that no significant particle contamination was induced by the process and that Ar/N2/F2 can be used as a highly clean and efficient etching gas.
Cleaning strategies for deposition chambers
CVD and ALD depositions usually cover the surface of the substrate and the chamber walls. Regular residual removal is necessary to obtain stable and repeatable deposition results with uniform surfaces at acceptable particle levels . Chamber cleaning gases such as, CF4 and C2F6, have been used for quite some time [2–3]. However, the strict requirements to sustain stable processes have led to more frequent chamber cleans. NF3 has emerged as the main gas for cleaning 300mm tools equipped with remote plasma source (RPS) systems; it shows significantly shorter clean times compared to CxFy–based cleans [5,6]. While NF3 meets the above mentioned requirements and is considered to be easily integrated into the fab, the gas is relatively expensive.
The Ar/N2/F2 used in this studied was delivered from conventional stainless steel cylinder gas bottles. Mixtures of 20% F2 are easily available for thermal cleaning processes . Our approach differs from on–site fluorine generation that utilizes hydrogen fluoride (HF) as feed material [8–10]. Since the deposition, etching, and selective removal of silicon–based dielectric materials  are very important process applications, we have used SiO2 and Si3N4 layers on Si for the tests.
The evaluation of the F2 gas mixtures was performed on 200mm wafers in an Applied Materials PECVD reactor on a P5000 main frame. The F2–containing gas bottle was installed inside a gas cabinet close to the CVD reactor. For the safe, clean and dry handling of the fluorine gases, a 3–way purging unit was part of the gas supply. The gas line was connected using a T–connection to the existing standard CF4 gas line, which is used as the standard Applied Materials chamber cleaning gas for this reactor. For precise gas flow control, a metal–sealed MFC was applied. Two F2 resistant O–rings (Isolast 9675) from Trelleborg Sealing Solution Germany GmbH were installed: one for the chamber lid outer seal, and one for the gas feed–through.
According to the Solvay–procedure regarding safety and stainless steel passivation with F2, a helium leak test, as well as a vacuum leak check for the relevant gas line, were done. N2F2 gas was then flowed for >10hrs to passivate the stainless steel surface.
Based on a former evaluation by Solvay, as well as Applied Materials’ BKM (best–known method) recipes, the first plasma ignition of N2/F2 gas was achieved immediately.
The best mixture and the best working parameters were selected calculating three L9–Taguchi matrices. The best mixture was identified to be the following composition: 10% Ar/ 70% N2/ 20%F2.
To investigate the etching gas performance, SiO2 and Si3N4 layers were used. Blank 200mm Si–wafers were deposited with a 1μm–thick oxide layer (silane–based PECVD); additionally, Si wafers were deposited with a 550nm–thick Si3N4 film.
As a complementary part of this work, the etch rate of the selected F2 mixture on amorphous silicon (a–Si) was measured. A layer of 300nm (the maximum allowed by this equipment) of a–Si was deposited on a layer of 100nm SiO2, which is the necessary substrate for thickness measurements.
To compare the etch performance of Ar/N2/F2 on SiO2 and Si3N4, respectively, tests using standard CVD cleaning gases, such as NF3, CF4 and C2F6/O2, were done. The CF4–based BKM chamber cleaning gas is recommended by Applied Materials. A further L9–Taguchi matrix was evaluated to define the best valid recipe for NF3.
Three repeatability runs with the selected 10%Ar/70%N2/20%F2 mixture were performed using a batch of 25 wafers/run. The wafers in slots 1, 12, and 25 were monitored for particle contamination. The particle performance behavior of the cleaning gas mixture was measured on particle monitor wafers by a Tencor Surfscan 6400. The minimal particle size measured was 0.25μm.
The first plasma ignition of the F2/N2 mixture could be achieved without observing arcing or other unknown effects, with the following parameters: Chamber pressure: 5Torr; N2/F2 gas flow rate of 1slm, temperature of 400 °C and 570 mils spacing.
Figure 1. Taguchi–L9 matrix of experiment of SiO2 etch rate (nm/min) ( ¦ ) and etch non uniformity (¿) (% 1σ) of N2/F2 gas as a function of a) flow rate, and b) pressure. The SiO2 layer thickness is 1μm.
The effects of the parameter variation on the performance of the F2/N2 plasma were evaluated through a first L–9 Taguchi matrix on SiO2. As seen in Fig. 1, there is a significant correspondence between the gas flow and chamber pressure to the achievable etch rate. The increase of the etch nonuniformity, is the limiting factor to improve etching performance.
Figure 2. Taguchi–L9 design of experiment of Si3N4 etch rate [nm/min] and etch non uniformity (% 1σ) of Ar/N2/F2 gas mixture as a function of a) Ar–flow, and b) pressure.
The cleaning gas behavior on Si3N4 with the addition of Ar at the N2/F2 mixture was studied next (Fig. 2). A preliminary Taguchi matrix study evidenced a reduced dependency of the etching performance on the RF power applied. A small addition of Ar notably increases the uniformity of the etching.
It is possible to achieve etching rates >1300nm/min even while maintaining nonuniformity values <5%, far below the value of 20%, which is generally considered the highest acceptable. These results are more encouraging because the addition of Ar to the F2/N2 mixture dilutes the concentration of the active specie F2, which is not desirable. Figure 2a shows the slight decrease in etch rate with increasing Ar–flow.
As a consequence of the described test results, the mixture 10% Ar / 70% N2 / 20% F2, was chosen for the next parts of this study. The mixture was premixed and delivered in a cylinder to avoid any F2 dilution effect by adding Ar during operation.
To test repeatability within wafer, wafer–to–wafer, and batch–to–batch of the Ar/N2/F2 etching and particle performance, three runs of 25 wafers each were performed. Processing the wafer batch was done in full automatic mode by the deposition of a 1μm SiO2 film for 35 sec. and a subsequent 60 sec. plasma chamber clean in Ar/N2/F2 gas. The particle monitor wafers were placed in slots 1, 12 and 25, respectively. Immediately after SiO2 deposition, the wafers left the CVD reaction chamber.
The cleaning recipe selected was a gas flow rate of 900 sccm Ar/N2/F2 and a chamber pressure of 2.1 Torr. The RF power was 800W and the susceptor temperature 400°C with 540 mils spacing (maximum spacing value on the silane–based oxide/nitride CVD chamber). The total cleaning time of 60 sec., included an over–cleaning of ∼30% to make sure any previously deposited oxide at the chamber walls, pumping plates, susceptor edges, and on the shower head will be completely removed.
The deposition recipe to deposit 1μm of SiO2 was the standard BKM recipe of Applied Materials with SiH4 and N2O as the main process gas.
Figure 3 shows the results of repeatability runs of 1μm SiO2 deposition followed by an Ar/N2/F2 chamber clean etch (run 2). The graphs display the oxide deposition thickness as well as the oxide deposition nonuniformity as a function of wafer position.
A SiO2 deposition of ∼1034nm was achieved for all 25 wafers. The oxide deposition nonuniformity was measured within–wafer as well as wafer–to–wafer. The within–wafer oxide deposition nonuniformity was ∼±1.4% (1σ) and wafer–to–wafer nonuniformity was ±0.8% (1σ), demonstrating a very stable oxide deposition.
To verify the repeatability of the SiO2 oxide etching rate for 25 wafers, the previously deposited wafers were run again and partially etched by Ar/N2/F2 for 30 sec. After measuring the remaining oxide thickness, the cleaning rate was calculated.
Figure 4. Repeatability (within–wafer and wafer–to–wafer) runs of SiO2 etching rate (nm/min) of Ar/N2/F2 for 30 sec. as a function of wafer position (total of 25 wafers).
Figure 4 shows the results of SiO2 etching rate of Ar/N2/F2 as a function of wafer position (run #2). For the first wafer, the etching rate was ∼1280nm/min and significantly lower compared to the rest of the 24 wafer batch, indicating the “first wafer effect.” From the second wafer on, the oxide etching rate averaged ∼1525nm/min for all wafers with a corresponding within–wafer oxide etch nonuniformity of ±7.1% (1σ). This shows, within wafer, as well as wafer–to–wafer, a very repeatable and stable oxide etching rate.
Summary of the three repeatability runs of 1μm– SiO2 deposition for 35 sec, followed by an Ar/N2/F2 chamber clean etching for ~60 sec.
To verify Ar/N2/F2 etch rate nonuniformity on SiO2 from batch–to–batch, three wafer runs (75 wafers) were carried out. Table 1 displays the result of all three runs. An average SiO2 deposition thickness of 1036nm was achieved for all three batches. For all three runs, the within–wafer oxide deposition nonuniformity averaged ±1.3% (1σ) and the wafer–to–wafer nonuniformity averaged ±0.7% (1σ), demonstrating a constant oxide deposition across all three batches. The data shows the average values of SiO2 layer deposition thickness (nm), oxide deposition nonuniformity (within wafer and wafer–to–wafer at %1σ) and Ar/N2/F2 oxide etching rate (nm/min) nonuniformity within wafer (%1σ) for three wafer runs (total 75 wafers). The average (mean) values of the batch–to–batch uniformity are shown as well. For the first run, the etch rate was not measured.
The Ar/N2/F2 etching rate on SiO2, batch–to–batch, was on average (mean) 1522 nm/min with a corresponding within–wafer oxide etch nonuniformity of ±8.0% (1σ). This demonstrates a very repeatable and stable oxide etching rate of the Ar/N2/F2 gas mixture for all wafer runs.
To get information about the particle performance behavior of the Ar/N2/F2 etching gas mixture on the wafer surface, the three particle monitor wafers were measured before and after the wafer run (Table 2). Using a Tencor Surfscan 6400, the minimal particle size measured was 0.25μm. The virgin reference wafer showed a particle contamination of ∼20 particles (0.25μm).
After SiO2 deposition and oxide etching, the average particle adders on the wafer surface was 22 particles for run 1, 14 particles for run 2, and 6 particles for run 3. On average (mean), just 14 particle adders were generated from run–to–run. The data indicate that no significant particle contamination was induced by the process.
All the obtained oxide etch rates and oxide etch nonuniformities of the Ar/N2/F2 cleaning gas mixture on SiO2 are equal to or better than the standard CF4 based BKM chamber clean of Applied Materials. The cleaning time achieved with the Ar/N2/F2 gas mixture was ∼30% shorter than the CF4–based BKM Applied Materials clean time. The particle values were <50 adders, which is well within specification for the AMAT P5000 process based on CF4 cleaning chemistry.
The chamber was opened after processing the first 25 wafer run, as well as after the final run (run 3), and visually inspected. For both inspections, the surface of the shower head, shadow ring, pumping plate and chamber lid O–rings, were clean and showed no signs of wear or residues.
Si–pieces with a thin SiO2 layer were placed on the pumping ring to check the cleaning efficiency outside the susceptor area. The complete oxide layer on the Si–pieces was removed throughout the three repeatability runs, which verifies that the Ar/N2/F2 gas mixture cleaned all important CVD chamber areas very well.
With the aim to achieve a direct performance comparison between the selected F2 mixture and NF3, a new L9 Taguchi matrix was calculated on SiO2 deposited layers to select the best working parameters for NF3 in the oxide/nitride chamber.
Amount of cleaning gas and cleaning time needed for three different cleaning gas chemistries (Ar/N2/F2, CF4, C2F6/O2, and NF3) normalized to 1μm dielectric film (SiO2 and Si3N4) thickness (20% over–cleaning time included).
NF3 flow rates >150sccm and a chamber pressure >0.6Torr cause nonuniformity >20%, which is not acceptable for a proper chamber cleaning. The maximum achievable etching rate with a nonuniformity <20% (1σ) was 1200nm/min. Table 3 summarizes all obtained etch rates for SiO2 and Si3N4.
To confirm the reliability of the obtained data on the oxide/nitride chamber, the etch rate on SiO2 and Si3N4 was measured in a tungsten chamber, where NF3 is a standard installation. For this chamber type, AMAT suggests a BKM chamber cleaning recipe. Its etching performance does not differ appreciably from the best parameter we set on the reference oxide/nitride chamber. It can be supposed that an F2 mixture can also improve cleaning performance in a TEOS chamber that normally uses a C2F6/O2/NF3 cleaning mixture, although no direct tests have been performed.
Comparison of the required F2 gas consumption [%] for Ar/N2/F2, CF4 and NF3 on SiO2 and Si3N4, respectively.
The total cost of a chamber cleaning process is determined from the wafer throughput of the CVD equipment and the cost of the consumables. Owing to the high cost of the equipment, the throughput is considered to have the highest impact. Using the etching rates of the different gas recipes in Table 4, it is possible to estimate the required time and amount of gas to etch, for example, 1µm of SiO2 and Si3N4. The required cleaning time is directly correlated to the throughput of the equipment, while the amount of required gas directly influences the cost of consumables.
The tested mixture Ar/N2/F2 showed improvements in both parameters, cleaning at a faster rate, and even requiring a lesser amount of gas. In particular, the NF3 cleaning method for SiO2 processes requires a longer cleaning time (+27%) and almost double the amount of process gas (+196%).
A layer of 300nm of a–Si was deposited on a 100nm layer of SiO2. Our standard Ar/N2/F2 mixture was applied for 15 sec., setting the plasma parameters at optimum conditions. The F2 cleaning recipe was allowed to etch both the a–Si layer and the SiO2 substrate completely and uniformly, resulting in an etching rate >1200nm/min. The test was repeated with the NF3 best cleaning recipe that was defined in this work. The a–Si layer was neither completely, nor uniformly etched.
The selected Ar/N2/F2 mixture can be applied as a drop–in replacement to the existing cleaning chemistries. The tests have been carried out on an industrial tool – an AMAT P5000 – which was demonstrated to be fully compatible with the new cleaning gas. The F2 gas line was connected to the existing CF4 gas line by applying a 3–way purging unit. The F2 mixture is compatible with stainless steel if a proper passivation procedure is performed.
The high etching rate and the low gas requirement of the selected Ar/N2/F2 mixture indicate that a significant cost–of–ownership reduction can be achieved. The cleaning recipes that have been taken as reference are a) the BKM CF4– cleaning recipe from AMAT and b) an NF3–based cleaning recipe that has been selected and optimized during the tests.
The global warming potential of the selected Ar/N2/F2 mixture is zero. This can contribute to the achievement of the greenhouse emissions reduction targets of the semiconductor industry. ¦
The authors would like to acknowledge Trelleborg Sealing Solutions GmbH (Stuttgart, Germany) for the delivery of F2 resistant O–ring seals (Isolast 9675).
- R. Hellriege, B. Hintze, H. Winzig, M. Albert, J.W. Bartha, et al., “Feasibility Study for Usage of Diluted Fluorine for Chamber Clean Etch Applications as an Environmental Friendly Replacement of NF3,” MRS, Materials Research Soc., Symp. Proc., Vol. 914, pp. 445, #0914–F09–23, April 2006.
- C. Allgood, “Delivering Fluorine to the Semiconductor Industry: Balancing Performance, Safety, Health and Environmental Issues,” Gases and Tech., p. 20 – 24, Sept./Oct. 2003.
- F. Blundo, M. Alberghina, M. Frazzica, “Selecting Proper Chamber–cleaning Processes for Installed–base CVD Tools,” IEEE/SEMI Adv. Semi. Mfg. conf., pp. 113–117, April 2005.
- W. F. Worth, “Achieving Sustainability in the Semiconductor Manufacturing Industry,” International SEMATECH Manufacturing Initiative (ISMI), Micro Magazine, vol. 24, no. 2, pp. 47–56, March 2006.
- O. F. Schedlbauer, H. F. Winzig, “Cost Reduction Challenges in CVD Chamber Cleaning: Strategies to Reduce Gas,” Future Fab Intl.., Vol. 13, Contamination control, pp. 164–169, July 2002.
- A. D. Johnson, W. R. Entley, P. J. Maroulis, “Reducing PFC Gas Emission from CVD Chamber Cleaning,” Solid State Tech., 43, Issue 12, p. 103, Dec. 2000.
- Y. Ohira, Y. Mitsui, T. Yonemura, K. Sakai, T. Takaichi, et al., “Evaluation of Fluorine Gas as an Alternative PECVD Chamber Cleaning Chemical,” Research Institute of Innovative Technology for the Earth (RITE) and National Institute of Advanced Industrial Science and Technology (AIST), June 2002.
- S. H. Siegele, F. J. Siegele, D. B. Hage, “On–Site Generation of High Purity Fluorine as a Safe and Economical Alternative for CVD Chamber Cleaning,” Future Fab. Intl. Vol. 13, Envir./ Health and Safety, pp. 107–116, July 2002.
- N. Irwin, G. Hodgson, B. Zievis, C. Myers, S. Orneals, et al., “Integrated Fluorine Supply Strategy for CVD Chamber Cleaning Applications,” SEMICON Southwest: A Partnership for PFC Emission Reductions Workshop, SEMI (ISBN # 1–882688–78–4), 2002.
- B. Jurcik, J. Sonobe, J–M. Girard, “Improving Thermal Cleaning Process,” Euro Asia Semi., pp. 15–18, April 2007.
- R. Wieland, Handbook of 3D Integration , Vol. 1, ch. 6, ,”SiO2,” edited by P. Garrou, C. Bower, P. Ramm, (978–3–527–32034–9), Sept. 2008.
Marcello Riva received his degree in chemical engineering at the Politecnico di Milano, Milan, Italy, and is technical sales manager at Solvay Fluor GmbH, Hans–Böckler–Allee 20, 30173 Hannover, Germany; ph.: +49 511 857 2648; email email@example.com.
Michael Pittroff received his degree in chemical engineering, Technische U. in Munich, Germany, and is a marketing manager at Solvay Fluor Korea.
Thomas Schwarze received his degree as technical assistant and is a product application specialist for inorganic fluorine compounds and fluorine specialties at Solvay Fluor GmbH.
John Oshinowo received his degree in physics at the Institute of Applied Physics, Hamburg U. (Germany) and his PhD in semiconductor technology at the Institute of Technical Physics – U. of Würzburg (Germany). As a technical consultant, he supports Solvay Fluor GmbH.
Robert Wieland received his masters in physical engineering, Ravensburg–Weingarten and is responsible for deposition and plasma dry etching in the 3D Integration group at the Fraunhofer Institute for Reliability and Microintegration (IZM), Hansastraße 27d, 80686 Munich, Germany.