Evaluating different approaches to critical HF-last cleaning
07/01/2003
Overview
While the basic chemistry used for cleaning silicon wafers still relies on HF and HCl, among others, increasingly the success of cleans for advanced processing lies in how these chemicals are controlled. One variation in control is in situ processing — doing etch, rinse, and dry in one tank; another is use of dilute chemicals. The first can be a key factor in eliminating particles; the latter can significantly lower defects.
From a comparison of a conventional multitank, HF-last process to a single-step, in situ HF-last process, we have found that dilute chemicals and in situ HF and drying are key factors in cleaning wafers with submicron critical dimensions. Our recent work has also shown that although numerous studies have been performed to analyze the mechanisms and kinetics of wafer cleaning processes, little attention, if any, has been given to monitoring and controlling chemical concentrations in process baths [1–4].
Increasingly, chemical concentration control is crucial to maintaining consistency and low cost of ownership, and to developing environmentally benign processes [1, 5]. Chemical concentration control becomes even more important as the composition of a cleaning bath becomes more dilute; tight control is necessary to maintain bath efficiency.
Our work was done with an Akrion GAMA automated wet station, which is capable of running multiple process sequences using multitank and single-tank in situ processes. Specifically, we compared a "conventional" multitank sequence of SC1, rinse, HF, rinse and dry to an "in situ" single-tank HF-HCl, rinse, and dry sequence. Our evaluation parameters were particle performance, metal signature, etch uniformity, and surface roughness.
We did our evaluations using a "pre-epi" HF-last clean because hydrogen-terminated silicon surfaces are some of the most critical cleans today. The epi growth process is very sensitive to wafer surface contamination. When done properly, this process yields a surface with the lowest number of defects. To simulate cleaning production patterned wafers, we alternated silicon wafers and dummy oxide-coated and etch wafers on the cleaning boat. This provided challenging contamination levels because dummy wafers contained etch byproducts that could transfer to bare silicon wafers.
For both multitank and single-tank cleaning, chemical concentration was controlled with an ICE-1 concentration control system; this system uses conductivity cells to provide fast, cost-effective and real-time chemical concentration control. Briefly explained, the conductivity sensors measure conductance by inducing an alternating current in the closed loop of the solution and measuring its magnitude. The current is proportional to the conductivity of the solution and is reported to an amplifier that scales the response to a 4–20mA output. Subsequently, the signal is fed into an analog module that scales it to µsec/cm, and then reports it directly to the GAMA computer. Amplifiers are calibrated with known conductivity standards.
Particle performance
For our particle evaluation, we first determined that the process steps and equipment were particle neutral; the average particle addition was -6 (1σ = 11). We found that the conventional HF-last process produced high particle counts (from 100 to almost 10,000 0.12µm particles). Consequently, the post-epitaxial defects were also high (>30,000). These light-point defects (LPDs) are considered to be poly-defects nucleated from silicate particles during the epitaxial deposition.
 Figure 1. Particle counts on 200mm bare silicon wafers, alternated with three types of oxide-coated etched wafers, after in situ hydrophobic clean. |
null
Our hypothesis was that conventional wafer transfer between tanks plays a significant role in increasing silicate deposition on silicon wafers. We believed that better particle performance could be achieved by minimizing or eliminating a wafer's exposure to the air-liquid interface, especially in the last step, where contaminants may reside. It is likely that these contaminants are deposited on wafers during insertion into or removal from the liquid.
 Figure 2. Oxide thickness from an ozonated rinse process. |
null
 |
null
When wafers were processed with the in situ clean with no transfer between cleaning steps, using a chemical injection scheme to process wafers in the dryer (i.e., by processing wafers to HF while submerged in the dryer), we obtained much lower particle counts (Fig. 1). Subsequently, the average LPD density/wafer was ~0.89 defects/cm2, significantly lower than any published data for 0.12&etam particles [6].
We also took measurements to characterize the residual oxide thickness after clean as a measure of the oxygen content on the wafer surface. This is important because the lower the oxygen content on an H-passivated silicon surface, the lower the post-epi defects. We found that for the ozone concentration used (i.e., 10ppm) oxide re-growth stabilized after a 5-min rinse; oxide growth was self-limiting and seemed to stabilize at ~8Å (Fig. 2).
Metals and oxide integrity
When using the in situ process, we found that the monitored metals (i.e., Al, Cu, Ni, Cr, Fe, and Zn) were <5x109 atoms/cm2 on H-terminated wafers. The advantage of using an HF-HCl mix is twofold: The HF removes the native oxide with metallic impurities, while HCl prevents the deposition of noble metal (e.g., Cu and dissolved Si) [7, 8]. Our data showed that much lower trace metal concentrations can be obtained with the in situ clean compared to that obtained when using the conventional SC2 clean. With SC2, the native oxide can have trace metals, such as Al, that can be hard to remove by HCl.
We also did gate oxide integrity testing to ensure a good Si-SiO2 interface after wafer cleaning. The potential drop (Vox) across the oxide film on silicon was also measured. Vox is most affected by metal contamination and electric charges located on the top oxide surface or near this surface. We found that Vox was higher than historical fab data collected over three weeks (Table 1). We also observed that the in situ clean resulted in higher yields; the number of failures on test devices was significantly less compared to historical fab data (Table 2).
Oxide etch
Oxide etching accuracy during wafer cleaning is strongly dependent on both bath temperature and tight control of HF concentration [9, 10]. Data gathered from our method of chemical concentration control show a linear relationship between HF concentration and solution conductivity over a wide range of HF concentrations. If temperature is held constant and the conductivity of the HF solution is maintained, the etch rate of SiO2 can be accurately controlled (Fig. 3).
We used a 100:1 HF solution at 27°C to study the performance of our method of chemical concentration control system. Over 8 hrs, we measured conductivity and HF concentration and determined etch rate every hour. We observed that conductivity remained steady and fluctuated only with subtle changes in fluoride concentration. This stability of HF concentration in the process bath was reflected in stable etch rates, which varied only ~1Å/min over 8 hrs. These data confirm the premise that control of HF concentration in the process tank by measuring conductivity will lead to stable SiO2 etch rates. Furthermore, concentration control is an effective method for controlling etch rates over a wide range of concentrations. A linear relationship between HF concentration (and conductivity) and etch rates can established. Thus, both dilute and concentrated HF processes can be regulated through concentration control.
 |
null
Our evaluation of SiO2 etch uniformity at dilute HF concentrations showed that conventional multitank etching provides more uniform etching. Once wafers are removed from the etch bath, the etch rate is reduced during transfer and eventually stops in the rinse tank. On the other hand, etch continues during an in situ rinse. This definitely has an adverse effect on etch uniformity due to uncontrolled fluid flow during the rinse.
 Figure 3. The addition of automatic concentration control for a 100:1 HF process at 25°C. |
null
The conclusion here is that well-designed etch and rinse baths are necessary and must be used to guarantee uniform and controlled etching. Although the conventional system demonstrates better uniformity, our results show that uniform etching can be obtained with a dilute chemical mix when combined with efficient rinsing.
Ismail Kashkoush, Gim Chen, Richard Novak, Akrion LLC, Allentown, Pennsylvania
Acknowledgments
GAMA and ICE-1 are trademarks of Akrion LLC.
References
1. I. Kashkoush, et al., Proceedings of the ECS Meeting and Conference, 4th Wafer Cleaning Symp., Chicago, IL, Oct 8–13, 1995, pp. 429–435.
2. M.J. Pelletier, et al., Semiconductor International, March 1996.
3. P.A. Giguere, H. J. Chen, Raman Spectr., 15, p. 199, 1984.
4. H. Kaigawa, et al., Journal of Applied Physics, 33, Part 2, 1994.
5. I. Kashkoush, et al., Materials Research Society Symposium Proc., Vol. 477, pp. 311–316, 1997.
6. S. Verhaverbeke, B. Pagliaro, Electrochemical Society Symp. Proceedings, Vol. 99–36, pp. 445–451.
7. C. Cowache, et al., Electrochemical Sociecy Symp. Proceedings, Vol. 99–36, pp. 59–68.
8. F. Tardif, et al., Electrochemical Society Symp. Proc., Vol. 95–26, pp. 159–160.
Ismail Kashkoush received his PhD from Clarkson University. He is director of applications and process engineering at Akrion LLC, 6330 Hedgewood Dr., Allentown, PA 18106; ph 610/530-3379, fax 610/530-3616, e-mail [email protected].
Gim Chen received his PhD from Rensselaer Polytechnic Institute. He is senior process engineer at Akrion.
Richard Novak received his PhD from the University of Illinois. He is VP of advanced technology and CTO at Akrion.