Characterizing Thermal Oxide Metals
Michael Radle Agilent Technologies, Bellevue, Wash. Huiling Lian, Beau Nicoley and Arnold J. Howard Intel Corp., Rio Rancho, N.M. -- Semiconductor International, 7/1/2001
The thermal oxidation of silicon to form a layer of silicon dioxide (SiO2) is an important step in semiconductor manufacturing. The SiO2 layer functions as a mask against impurity diffusion or ion implantation, it serves as a dielectric layer in MOS devices confining charge carrier flow, and it isolates closely spaced devices from each other. Silicon wafers, as shipped from the manufacturer, have a native oxide layer ~30-50 Å thick. A device manufacturer requires a much thicker oxide layer to obtain the physical and electrical properties described above. Exposure to oxygen (dry technique), or water vapor (wet technique) at elevated temperatures allows the growth of silicon oxide layers 100-20,000 Å thick. Impurities or contamination in the thermal oxide will have a deleterious effect on successful IC manufacturing.SME combined with ICP-MS can provide valuable information on the type, source, and levels of metallic contamination in several semiconductor processing steps. Surface contaminant concentration techniques have been applied to determine native and thermal oxide layer purity on silicon wafers.1-4 Advantages of SME-ICP-MS include accurate analysis of up to 40 elements in a single droplet, excellent ultra-trace detection limits (parts per trillion), spot or entire wafer surface contamination information, and 60 min turnaround time for wafer preparation and ICP-MS analysis. In addition, improved accuracy and precision can be realized through the use of matrix-matched standards. These features of the SME-ICP-MS technique make it uniquely suited to perform real-time wafer production monitoring.
Surface extraction chemistry
The surface extraction is designed to remove the SiO2 layer that resides on top of the silicon crystal substrate. Any metallic contaminants on or within the oxide layer will also be dissolved into the extraction droplet, as will any metallic contamination on the surface of the silicon crystal substrate.
The primary chemical reaction occurs in an exposure chamber filled with hydrofluoric acid vapor. It involves the decomposition of the oxide layer by hydrofluoric acid vapor:
SiO2(s) + 4HF(aq) ® SiF4(aq) + 2H2O(l) (1)
Metallic contaminants on the wafer surface undergo the following reactions to form soluble metal fluorides:
CuO(s) + 2HF(aq) ® CuF2(aq) + H2O(l) (2)
Unless the appropriate precautions are taken, metals that are more electronegative than silicon (i.e. Au and Cu) will react further to redeposit on the wafer surface. The redeposition can occur through both electrochemical and galvanic reactions to form the metal:
2CuF2(aq) + Si(s) ® SiF4(aq) + 2Cu(s) (Electrochemical)
Cu+2(aq) + 2e- ® Cu(s) (Galvanic)
To minimize the above reactions, the extraction droplet chemistry is kept highly oxidative, preventing redeposition of metal contaminants onto the wafer surface. Nitric acid and hydrogen peroxide were tested as the oxidative component of the extraction droplet in a previous study (Table 1).4 Copper was used for this comparison because it is a common contaminant, and has shown to be difficult to recover from the wafer surface. Droplets containing H2O2 resulted in minimal deposition losses due to reaction of the metal contaminants to form soluble metal fluorides:
Cu(s) + 2HF(aq) + H2O2(aq) ® CuF2(aq) + 2H2O(l) (3)
| Table 1. Effect of Droplet Chemical Composition on Copper Recovery | ||
| Formulation | Droplet composition (%)* | % Recovery of copper |
| #1 | 0.5 HNO3, 5 HF | 37 |
| #2 | 6.0 H2O2, 5 HF | 86 |
SME sample preparation
Silicon wafers were placed in a cleaned PTFE chamber containing HF vapor, with an exposure time of 30 min. The HF vapor reacted with the SiO2 layer on the wafer surface, causing it to dissolve. As the layer dissolved, micro-droplets containing the products of Reaction 1 formed on the wafer surface. The micro-droplets were composed of soluble silicon fluoride, water and metallic contaminants from the SiO2 layer and Si substrate surface. From Reaction 1, it can be seen that the amount of water formed during the dissolution process will be proportional to the thickness of the SiO2 layer. The additional water formed on the wafer surface from the chemical reaction was accounted for by the final extraction droplet volume term used in the conversion of contaminant concentration to atoms/cm2.
After the dissolution was complete, the surface was manually scanned with a 250 µL extraction droplet, which collects the micro-droplets as it is scanned across the wafer surface. Once the entire surface has been scanned, the extraction droplet is pipetted from the wafer surface into a PTFE vial. The scanning process is repeated with another 250 µL extraction droplet, which is then added to the first droplet. This results in a final droplet volume of ~0.65 mL, which is then diluted to a final volume of 0.75 mL with DI water.
ICP-MS analysis
An Agilent 4500 ICP-MS was used for the trace metal analysis using both cool and normal plasma modes of operation. The ICP-MS was equipped with a cross-flow nebulizer specifically designed for use with limited sample volumes. The sample was self-aspirated at an uptake rate of 100 µL/min to accommodate the small extraction droplet volume. Platinum interface cones were used for the analysis. Instrument operating conditions for cool plasma and normal plasma analysis are given in Table 2.
| Table 2. ICP-MS Operating Conditions for SME Analysis | ||
| Parameter | Cool plasma analysis | Normal plasma analysis |
| ICP forward power (W) | 930 | 1250 |
| Nebulizer gas flow rate (L/min) | 0.91 | 1.04 |
| Blend gas (L/min) | 0.76 | 0.06 |
| Spray chamber temperature (°C) | 2 | 2 |
| Sampling depth (mm) | 17.3 | 7.9 |
| Data acquisition parameters | 3 points per mass, | 3 points per mass, |
| 1 sec per point | 1 sec per point | |
Matrix effects, potential interferences
There are two potential sources of physical interferences that may result from the sample matrix. The first is the presence of dissolved silicon from the wafer's SiO2 layer, and the second is due to the nature and concentration of reagents used in the extraction process, i.e. 5% HF and 6% H2O2. These matrix components can give rise to differences in nebulization and sample transport efficiency between the extraction droplet and the aqueous standards used to calibrate the ICP-MS instrument. The higher concentrations of silicon and oxidizing reagents can also result in physical changes in the plasma (i.e. plasma loading).
A sample containing a significant concentration of solids or chemical matrix absorbs a larger amount of ionizing energy from the plasma than does an aqueous calibration standard. If the energy consumption is high enough, it can suppress signals for difficult-to-ionize elements such as B, Zn and Cd. Judicious choice of instrument operating parameters and an appropriate calibration scheme will minimize the effect of the interferences on the measurement.
The analysis of critical elements in semiconductor process monitoring requires the use of cool plasma conditions. Cool plasma conditions are characterized by operation of the plasma at reduced forward power, in conjunction with the ShieldTorch interface to virtually eliminate the formation of polyatomic interferences. The resulting change in plasma potential requires optimization of the sampling depth, carrier/blend gas flows, and ion lens settings in order to fully realize the benefits of this technique. Cool plasma analysis for critical semiconductor elements such as Ca, K and Fe has historically been performed at 600-800 W, which, along with reducing argon-based polyatomic interferences, also significantly reduces the ionization potential of the plasma. This can result in the loss of sensitivity for elements with high ionization potentials such as B, Zn and Cd. These low-power cool plasma conditions can also result in significant plasma loading effects when analyzing high matrix samples, such as wafer extractions containing high concentrations of silicon and other chemical reagents.
The SME-ICP-MS results reported here were performed using a plasma power of 930 W. The higher plasma power compensates for plasma loading caused by dissolved solids and reagents in the VPD sample matrix while reducing argon-based interferences on critical elements such as Fe and Ca. It should also be noted that, at this power level, silicon and fluorine are not ionized, thereby preventing the formation of their molecular ion interferences.
Physical interferences associated with the chemical composition of the VPD extraction matrix can also be compensated for with matrix-matched calibration standards. The calibration blank and standards used in this study were prepared at 0, 100, 250 and 500 ppt in a 5% HF, 6% H2O2 matrix.
Evaluating interference reduction
Matrix effects from the extraction sample matrix were evaluated to determine the extent of sample nebulization, transport and plasma loading interferences. The results were also used to identify at what concentration of silicon suppression of analytical signal begins. This will define the maximum permissible silicon concentration for SME sample analysis.
| Cool plasma spike recoveries from 100, 250, 500 and 1000 ppm silicon extraction matrix solutions. |
The data exhibit excellent spike recoveries for the extraction matrix containing 100 ppm silicon. This indicates minimal interference effects from the chemical extraction matrix and silicon at these concentrations. The spike recovery data for the higher-level silicon solutions show signs of analyte signal suppression resulting in lower spike recoveries. An interesting observation regarding the spike recovery data is that, in each silicon solution, the elemental recoveries appear in order of their first ionization energy. It can be observed that the element with the highest first ionization energy has the lowest spike recovery in each of the silicon solutions. This indicates the presence of a reduction of the effective ionization energy in the plasma due to the chemical reagents and silicon in the extraction matrix (i.e. plasma loading). The exception to this observation is aluminum in the 250 ppm silicon solution, which has a higher recovery than expected. Further inspection of the data for the 250 ppm silicon matrix reveals the same inconsistency for sodium, indicating probable contamination with both of the elements.
In addition to plasma loading, there is a nebulization efficiency interference caused by increased levels of dissolved silicon in the samples. This can be seen in the spike recovery patterns as the silicon concentration is increased from 100 to 1000 ppm. The successful recovery of the spikes in 100 ppm silicon indicates that the use of higher plasma power (930 W) and matrix-matched standards are effective in overcoming the physical interferences associated with this type of sample. Cool and normal spike recovery data for the 100 ppm silicon extraction matrix are given in Tables 3 and 4, respectively. The data show excellent spike recoveries, indicating negligible interference effect on plasma ionization efficiency due to the extraction sample matrix.
| Table 3. Cool Plasma Spike Recovery Data | |||
| Element | Mass | Spike concentration (495 ppt) | % Recovery |
| Lithium | 7 | 495 | 100.9 |
| Sodium | 23 | 495 | 98.0 |
| Magnesium | 24 | 495 | 96.8 |
| Aluminum | 27 | 495 | 99.4 |
| Potassium | 39 | 495 | 97.2 |
| Calcium | 40 | 495 | 96.7 |
| Chromium | 52 | 495 | 97.4 |
| Manganese | 55 | 495 | 94.7 |
| Iron | 56 | 495 | 86.6 |
| Iron | 57 | 495 | 89.7 |
| Nickel | 58 | 495 | 97.0 |
| Nickel | 60 | 495 | 94.4 |
| Copper | 65 | 495 | 99.2 |
| Tin | 118 | 495 | 95.5 |
| Table 4. Normal Plasma Spike Recovery Data | |||
| Element | Mass | Spike concentration (495 ppt) | % Recovery |
| Boron | 11 | 495 | 82.1 |
| Titanium* | 48 | 495 | 101.8 |
| Chromium | 52 | 495 | 81.5 |
| Manganese | 55 | 495 | 99.0 |
| Nickel | 60 | 495 | 92.5 |
| Zinc* | 68 | 495 | 127.3 |
| Indium | 115 | 495 | 88.2 |
| Tin | 118 | 495 | 96.6 |
| Barium | 138 | 495 | 99.6 |
| Lead | 208 | 495 | 99.8 |
| *Matrix interference present, see Table 5. | |||
| Table 5. Molecular Ion Interferences Found in SME Matrix | ||
| Analytical isotope | Molecular ion interference | Recommendation |
| 44Ca | 28Si16O | Use 40Ca |
| 47Ti | 28Si19F | Matrix elimination |
| 48Ti | 28Si19F1H | Matrix elimination |
| 49Ti | 30Si19F | Matrix elimination |
| 59Co | 40Ar19F | Cool plasma |
| 63Cu | 28Si19F16O | Use 63Cu, cool plasma |
| 66Zn | 38Ar28Si | Use 68Zn, cool plasma |
| 68Zn | 40Ar28Si | Use 68Zn, cool plasma |
ICP-MS detection limits
Table 6 lists the aqueous detection limits determined in the extraction droplet chemical matrix containing 100 ppm Si. The 3 s detection limit is calculated using the 100 ppm silicon blank, and the 100 ppm silicon with the 495 ppt spike used in the silicon suppression study shown in Figure 1:
Detection limit (ppt) = (3× std. dev. of blank counts)/(standard counts-blank counts)× (concentration of standard in ppt)
| Table 6. Agilent 4500 ICP-MS Solution and Wafer Detection Limit Data | ||||
| Element | Aqueous 3 s DL (ppt) | 200 mm wafer DL, 750 µL drop (atoms/cm2) | 300 mm wafer DL, 750 µL drop (atoms/cm2) | ITRS 300 mm wafer requirements, 1999 (atoms/cm2) |
| Li | 0.05 | 1.0 × 107 | 4.6 × 106 | <1.0 × 1011 |
| B | 15 | 2.0 × 109 | 8.7 × 108 | Dopant, no spec. |
| Na | 6 | 3.1 × 108 | 1.7 × 108 | <1.0 × 1011 |
| Mg | 2 | 1.2 × 108 | 5.3 × 107 | <1.0 × 1011 |
| Al | 1 | 5.3 × 107 | 2.4 × 107 | <1.0 × 1011 |
| K | 3 | 1.1 × 108 | 4.9 × 107 | <1.0 × 1011 |
| Ca | 8 | 2.9 × 108 | 1.3 × 108 | <1.3 × 1010 |
| Cr | 7 | 1.9 × 108 | 8.6 × 107 | <1.3 × 1010 |
| Mn | 0.4 | 1.0 × 107 | 4.6 × 106 | <1.3 × 1010 |
| Fe | 14 | 3.6 × 108 | 1.6 × 108 | <1.3 × 1010 |
| Ni | 11 | 2.6 × 108 | 1.2 × 108 | <1.3 × 1010 |
| Cu | 8 | 1.8 × 108 | 7.9 × 107 | <1.3 × 1010 |
| Sn | 47 | 5.7 × 108 | 2.6 × 108 | <1.0 × 1011 |
| Ba | 0.2 | 2.1 × 106 | 9.3 × 105 | <1.0 × 1011 |
| Pb | 0.4 | 2.8 × 106 | 1.2 × 106 | <1.0 × 1011 |
This data was used to calculate corresponding wafer detection limits (atoms/cm2) for surface metals on 200 and 300 mm wafers. Also listed in Table 6 are the 1999 ITRS 300 mm wafer surface metal requirements.6 The SME process employed, combined with metals determination using the Agilent 4500 ICP-MS, exceeds all required detection levels for surface metal contaminants by several orders of magnitude.
Conclusion
The SME-ICP-MS technique offers a sensitive and accurate method for characterizing trace metals on silicon wafer surfaces. Silicon wafers can be prepared and analyzed in less than an hour, providing real-time data for manufacturing quality assessment. The chemical composition of the extraction droplet in the SME process has been shown to be critical to ensuring efficient extraction of metal contaminants from the wafer surface. The use of an HF/H2O2 formulation has been shown to provide excellent extraction efficiencies.
Potential physical interferences associated with analysis of the SME droplet matrix by ICP-MS can be virtually eliminated. This is achieved through the use of matrix-matched calibration standards and through judicious selection of instrument operating conditions. The operation of the ICP-MS at a plasma power higher than that typically used in cool plasma analysis is critical for the elimination of physical interferences arising from the SME sample matrix.
Michael Radle is an ICP-MS applications engineer for Agilent Technologies in Portland, Ore., and is responsible for semiconductor industry ICP-MS support in North America. He has 16 years of ICP-AES and ICP-MS analytical experience in the environmental, nuclear and semiconductor industries. He has a B.S. in chemistry from the University of Wisconsin.Huiling Lian was an analytical process engineer at Intel's New Mexico Materials Lab, responsible for semiconductor chemical analysis and process improvement. She has a Ph.D. in chemistry from the University of Texas.
Beau Nicoley is a chemical analysis technician for Intel, responsible for chemical purity analysis for the New Mexico site. He has six years of analytical experience, three of which have been focused on ICP-MS, GFAA and VPD.
Arnold Howard is department manager for Intel's manufacturing quality analytical group. He was previously manager of the New Mexico Materials Lab, and a metrology group leader/litho process development engineer for flash memory products. Prior to joining Intel, he worked for five years at Sandia National Labs (Albuquerque, N.M.) on GaAs opto/microelectronic process development, integration and characterization. He has a Ph.D. in chemical engineering from the University of Florida, and bachelor's and master's degrees in chemical engineering from the University of Pennsylvania and the University of Virginia, respectively.
REFERENCES
- A. Shimaki, H. Hiratsuka, Y. Matsushita, S. Yoshii, "Chemical Analysis of Ultratrace Impurities in SiO2 Films," Extended Abstracts of 16th Int. Conf. on Solid State Devices and Materials, Kobe, Japan, 1984, p. 281.
- L. Hall, J. Sees, B. Schmidt, "Characterization and Application of the Vapor Phase Decomposition Technique for Trace Metals Analysis on Silicon Oxide Surfaces," Surface and Interface Analysis, 1996, Vol. 4, p. 511.
- S. Tan, "Application of Vapor Phase Decomposition Techniques for Trace Metals Analysis in Oxide Coatings on Silicon," Nuclear Instruments and Methods in Physics Research, 1995, p. 458.
- F. Meyer, J.B. White, "Acid Etching Chemistry Characterizes Silicon Wafer Surface Metals," Semiconductor International, July 1999, p. 137.
- D. Burkman, D. Deal, D. Grant, C. Perterson, "Aqueous Cleaning Processes," Handbook of Semiconductor Wafer Cleaning Technology, 1993, p. 111.
- The National Technology Roadmap for Semiconductors, San Jose, Semiconductor Industry Association, 1994.
