EXTREME SILICON

Using ICP-MS for in-line monitoring of metallics in silicon wafer—cleaning baths

Larry W. Shive, Kenneth Ruth, and Philip Schmidt, MEMC Electronic Materials

A setup at a wafer manufacturing facility connects an in-line ICP-MS to final cleaning baths for silicon wafers to monitor metal concentrations in all the baths as the wafers are cleaned.

Silicon wafer manufacturers con-tinue to improve processes and products through design of experiments approaches and the implementation of statistical process control (SPC). Although SPC may be established by monitoring flatness, surface metals, or other output parameters, controlling such critical input parameters as polishing pressure or bath temperature is preferable. SPC of surface metals on silicon wafer surfaces is typically accomplished by controlling several inputs, such as bath temperature, bath concentration of major components, bath life, and feed chemical purity, as well as by monitoring one output—surface metals. This control system has several weaknesses. The actual purity of each chemical batch is not known, although it is inferred from the general purity specification of the raw materials. The cost to sample wafers from each cleaning lot is unacceptably high, and therefore the wafer sampling rate is typically reduced to a few wafers per day. As a result, process upsets may be missed or go undetected for several hours.

Such weaknesses may be overcome by frequent in-line monitoring of the metals levels in every cleaning bath or, at least, in the baths that are critical for each metal. This has been done by coupling in-line sampling equipment with standard inductively coupled plasma mass spectrometry (ICP-MS) analytical tools. In-line sampling is essential to the success of this approach because it minimizes contamination while maximizing the possible sampling rate. It is very difficult to manually sample a chemical bath with purity levels of <1 ppb without significantly contaminating the sample, and high-frequency manual sampling of baths is very labor intensive.

A key assumption behind all of this work is that there is a mathematical relationship between the amount of a specific metal on a wafer and the amount of metal in the last cleaning solution in which the wafer was immersed. This equation may include immersion time, chemical concentration, bath temperature, and the amount of metal on the wafer before immersion. A second assumption is that a linear combination of baths may be modeled by a linear combination of these mathematical relationships. Therefore, complete knowledge of these equations; bath temperatures, times, and concentrations; and incoming wafer quality is sufficient to predict the metal density on a wafer at any point in the process. Several investigations have already established mathematical models to predict surface metals as a function of bath metal concentration after key cleaning steps.^1–3

This article discusses a setup used at the MEMC Electronic Materials wafer manufacturing facility (St. Peters, MO), where an in-line ICP-MS is connected to final cleaning baths for silicon wafers in order to monitor metal concentrations in all the baths as the wafers are cleaned. Use of the analytical equipment and process has switched the fab's mode of operation from product output monitoring to process input monitoring, ensuring a more consistent product and a level of process control consistent with the metals levels needed for 300-mm wafers, as detailed in the SIA roadmap and related industry documents.

Experimental Procedures

Figure 1 is a simple schematic of the analysis system connected to a chemical wet bench. Multiple baths are connected to the same collection point. All wetted parts are made of either Teflon, PVDF, polyethylene, or quartz. This same scheme is used to connect with and sample all the baths in a cleaner. The components of the system are:

• A chemical cleaning or rinse bath.

• A recirculation or micrometering pump.

• A sampling line connected to the recirculation line on the outlet side of the pump or to the micrometering pump.

• A collection point.

• An autoinjector.

• An ICP-MS.

A Model 4500 ICP-MS (Hewlett-Packard, Palo Alto, CA) with a crossflow nebulizer was used for this work. The instrument was operated under cool plasma conditions, which means that the HP shield torch system is being used. This system minimizes such polyatomic ion interferences as ArH (amu = 39) and ArO (amu = 56). With these ion interferences reduced, the instrument can detect potassium, calcium, and iron at ultratrace levels. Detection limits for selected metals in a nitric acid matrix are listed in Table I. A 1000-ppt NIST-traceable standard was used to prepare 100-ppt and 10-ppt solutions. These three solutions, plus a blank, were used for calibration every 16—24 hrs. Drift of <5% was confirmed hourly using a 100-ppt NIST-traceable standard. The data collected included cleaning tool identification, bath identification, sampling date, sampling time, and individual metal concentration. The individual bath data were stored in a manufacturing database and reviewed by the engineering staff. Wafer surface metal analysis was done using acid-drop extraction with the extracted droplet examined by the ICP-MS, a method described in detail elsewhere.^4,5 Cold plasma ICP-MS analysis of typical silicon wafer—cleaning chemicals has also been explained elsewhere.⁶

Na (23)	Mg (24)	Al (27)	K (39)	Ca (40)	Cr (52)
2.90	2.27	2.17	4.00	2.74	2.39
Mn (55)	Fe (56)	Co (59)	Ni (60)	Cu (65)	Zn (66)
5.44	2.77	2.99	5.00	5.19	12.54

Table I: Typical detection limits in parts per trillion for selected metal isotopes by ICP-MS. The atomic mass units of the isotopes are given in parentheses.

Results and Discussion

All metals measured in the rinse baths were either found to be below the detection limit or matched the expected process capability of the water purification process. Therefore, it was concluded that the sampling method is essentially noncontaminating down to the detection limits of the ICP-MS or the process capability of the water purification system.

Figure 2 shows the trend of metals concentration in an SC-1 bath over 1 month. In this particular process tank, most of the metals were <100 ppt. Several metals were consistently found to be at or below their detection limits and, as a result, they were sometimes determined to be present in negative concentrations. Concentrations were calculated to be negative whenever the bath sample had a lower concentration than the blank. In order to correctly analyze the data, statistical analysis of all the data, including the negative numbers, was always performed.

Figure 3 illustrates the day-to-day variation of sodium in a cleaning bath and on wafers processed in that bath. In this particular bath, the variation range of the sodium in the bath was >100x. Also, after a shift of concentration occurred, the metal tended to stay at that level for several days before shifting again. Sodium density on the wafer surface mirrored these shifts in the bath. In this actual situation, the exact date and time of the changes in sodium from this database were used to identify a specific piece of equipment in the process that was the source of these changes. The data from this trend chart were analyzed to determine whether there was a relationship between sodium in the bath and on the wafers. A simple linear regression analysis of the data plotted in Figure 4 indicated that there was a correlation. In such an example, this correlation can be used to set an upper control limit for sodium in this cleaning bath and, if this limit is exceeded, wafer processing can be stopped and corrective actions taken to ensure that no defective wafers are produced.

Bath lifetimes can be defined logically and systematically based on in-line ICP-MS data. In many manufacturing facilities, bath lifetimes are arbitrarily based on a conservative approach that is supposed to guarantee product quality or are tied to manufacturing personnel shift changes or to the total number of wafers processed. For example, baths may be changed every 4, 8, or 12 hours. Arbitrary bath changes may waste machine time and chemicals and may not necessarily guarantee product quality. Off-line methods for defining SC-1 bath life as well as an in-line optical method for monitoring copper deposition from HF have both been discussed recently.^6,7

Figure 5 depicts the concentration trends of aluminum and iron in a chemical bath, calculated as a function of bath life with a constant feed of wafers. Several hundred wafers were typically cleaned in this bath during one bath life. A steady increase in metal levels over time after a new bath was made can be observed. A maximum bath life may be chosen using these data in conjunction with the correlation of bath purity to wafer purity.

In some instances, it may not be possible to establish a relationship between bath and wafer purities. Two possible reasons for this are the limits of detection, either in the bath or on the wafer, and the chemistry of the metal in the bath, such that it does not deposit at any concentration. Figure 6 shows one such example, where the detection limit of iron in the bath was not low enough to allow us to detect the iron in the bath and observe the relationship that exists at higher concentrations. In this case, parts-per-quadrillion detection limits would be required to perform the analysis.

Conclusion

The in-line ICP-MS method for bath analysis has been demonstrated as a tool to monitor concentrations of metals in cleaning baths, a critical process input parameter. Detection limits of <10 ppt have been achieved for water as well as for typical SC-1 and SC-2 bath chemistries. The method does not contaminate the process fluid being analyzed, and it can be fully automated to sample each of the process baths in a cleaning line with a frequency determined by the user. The data obtained from this analytical system can be used to successfully identify shifts in process fluid purity and to track them to a specific cleaner, time, cleaning bath, raw material, or specific piece of equipment in the process. In addition, mathematical relationships between bath purity and wafer purity can be established. From these, upper control limits can be set for specific metals in specific cleaning baths, so that when they are exceeded the process is stopped and all material cleaned during the upset can be immediately contained. An added benefit of this method is that bath lifetimes may be set more logically based on the desired quality of the product and the correlation with bath purity. In some cases, bath life may be extended considerably, thereby reducing costs and increasing the on-line time of the tool. Finally, whenever the concentration of some metals is consistently <10 ppt in a specific bath or below 1 x 10⁹ atom/cm² on the wafer surface, bath-to-wafer purity relationships can be difficult to establish. In these cases, detection limits must be extended to 1 ppq in the baths and 1 x 10⁷ atom/cm² on wafers to define these equations.

References

1. G Maeda et al., "Adsorption and Desorption of Contaminant Metals on Si Wafer Surface in SC-1 Solution," in Ultraclean Semiconductor Processing Technology and Surface Chemical Cleaning and Passivation (Pittsburgh: Materials Research Society, 1995), vol. 386, 195–
206.

2. S Dhanda et al., "Iron Deposition from SC-1 on Silicon Wafer Surfaces," in Ultraclean Semiconductor Processing Technology and Surface Chemical Cleaning and Passivation (Pittsburgh: Materials Research Society, 1995), vol. 386, 201–206.

3. I Teerlinck et al., "Effect of Anions on Copper Outplating from HF Solutions," in Ultra Clean Processing of Silicon Surfaces—UCPSS '96 (Leuven, Belgium: IMEC, 1996), 21–24.

4. F Meyer, JB White, and M Radle, "Silicon Wafer Surface Metals Characterization Using Automatic Wafer Scanning and Inductively Coupled Plasma Mass Spectrometry," to be published in Semiconductor International (1999).

5. G Settembre and E Debrah, "Using VPD ICP-MS to Monitor Trace Metals on Unpatterned Wafer Surfaces," MICRO 16, no. 6 (1998): 79–89.

6. R Mortensen and T Gluodenis, "Cold Plasma Extends Trace Metal Detection Capability," Semiconductor International 21, no. 7 (1998): 261–266.

7. D Chopra and II Suni, "An Optical Method for Monitoring Metal Contamination during Aqueous Processing of Silicon Wafers," Journal of the Electrochemical Society 145, no. 5 (1998): 1688–1692.

Larry W. Shive, PhD, is an MEMC science fellow and has been manager of cleaning, inspection, and packaging R&D in the technology department of MEMC Electronic Materials (St. Peters, MO) for eight years. He has a PhD in chemistry from Texas A&M University. (Shive can be reached at 314/279-5370 or lshive@memc.com.)

Kenneth Ruth is a senior research technician supervisor at MEMC Electronic Materials. He has 10 years of experience with development and application of trace metal analytical techniques. His current responsibilities include first-line management, development, and data automation of surface analytical resources.

Philip Schmidt is a cleaning process engineer in the manufacturing technology department of MEMC Electronic Materials. He holds a BS in industrial technology from Southeast Missouri State University.

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit