Catching the Demon of the Semiconductor Process－Wafer Surface Contamination

The cleanliness of a wafer’s surface will affect the subsequent semiconductor processes and product yield to a certain extent. The most common contaminants include metals, organic residue and granular particles. Contamination analysis results can reflect upon a certain process step, the degree and type of contamination encountered in a specific machine or in the overall process. Early literature pointed out that losses due to failure to effectively remove contamination from wafer surfaces during the manufacturing process may account for more than 50% of all yield losses.

Common contaminants can have many possible impacts on the process and product. For example, metal contamination causes leakage currents in p-n structures. This in turn leads to a decrease in the breakdown voltage of the oxide and a reduction in the carrier lifetime. Organic contaminants can lead to the development of unexpected hydrophobic properties on the wafer’s surface, an increase of the surface roughness, and the generation of a surface haze as well as the destruction of epitaxial growth. Furthermore, if the contaminants are not removed first, they can also affect the effectiveness of the cleaning of metal contamination. Then there is particle contamination, which can cause problems during etching and lithography, producing blocking or masking effects. During film growth or deposition, they can generate pinholes and microvoids. If the particles are large and conductive, they can even cause short-circuiting.

Therefore, how to ensure that there is no contamination on a wafer’s surface has always been an important issue. After the introduction of RCA mixture cleaning solutions based on hydrogen peroxide in the 1970s, many different cleaning formulas have been used. For example, there is the traditional, two-step cleaning process that begins with the use of a 1:1:5 ratio ammonia water:hydrogen peroxide:pure water (SC-1) solution followed by a 1:1:6 ratio of hydrochloric acid:hydrogen peroxide:pure water (SC-2) [1]. The efficacy of different cleaning processes depends on your needs and objectives. They should also be evaluated and tested according to the specific machines being used. The hope is that we can clean and remove the contamination on wafer surfaces by introducing different cleaning processes.

Many analytical techniques and instruments have been used for the detection of metal contamination in the past, including Auger Electron Spectroscopy (AES), Secondary Ion Mass Spectrometry (SIMS), Time of Flight-Secondary Ion Mass Spectrometry (TOF-SIMS), Rutherford Backscattering Spectrometry (RBS) and Graphite Furnace Atomic Absorption Spectrometry (GF-AAS), etc. However, each tool or technique has its own limitations. For example, insufficient sensitivity results in the inability to provide accurate quantitative results, meaning that the analysis conducted would be superficial. Other limitations include the inability to detect and analyze multiple elements at the same time and lengthy overall measurement times. In other words, all of the above analysis techniques have limited analytical capabilities.

In the past, the concentration detection requirement for metal on a wafer’s surface was 10¹⁰ atoms/cm². As processes evolved, the detection limit was reduced to 10⁸ atoms/cm². The main technologies able to meet this analysis requirement are Total Reflection X-ray Fluorescence (TXRF) and ICP-MS. So how do you choose between these two types of testing instruments to make sure that you get the most out of your metal contamination analysis?

TXRF has the advantages of being nondestructive and able to perform fixed-point analysis. In the detection of transition metal elements such as copper and iron, its detection limit is about 10⁹–10¹⁰ atoms/cm². As such, it can meet general testing and monitoring needs. For metals with low atomic masses like sodium, magnesium, and aluminum, however, there is a phenomenon where the detection limit is too high. Furthermore, lithium, beryllium, and boron cannot be detected. It has also been pointed out in the literature that TXRF has a low recovery rate for the detection of copper elements. TXRF enables direct measurement of samples or, when used in combination with the rolling droplet method and the Vapor Phase Decomposition (VPD) sample preparation method, the conducting of surface contamination analysis.

The principle behind this type of analysis is to irradiate a sample surface using a monochromatic X-ray source at an angle that is less than the total reflection angle. This excites the sample’s surface within a thickness of only 3-8nm. The resulting fluorescence is detected by a detector positioned perpendicular to the sample’s surface, which provides qualitative information by analyzing the energy wavelengths. The signal strength can be converted into quantitative information via the calibration curve.

To increase the detection capability of the instrument to the detection limit of 10⁸–10⁹ atoms/cm², consider using it in combination with sample pre-concentration via VPD or Vapor Phase Treatment (VPT) [2-4]. You can also use a Synchrotron Radiation (SR) light source to enhance the intensity of the incident light and improve detection ability [5]. When used in tandem with VPD or VPT systems, whether it is a commercial machine or a self-designed and assembled system, the important thing is to place the wafer in a high-cleanliness environment. Then hydrofluoric acid vapor is introduced. It condenses on the surface layer of hydrophilic silicon oxide, and the silicon oxide is decomposed according to the following formula.

SiO₂ + 6HF → H₂SiF₆ + 2H₂O

H₂SiF₆ → SiF₄ + 2HF

Droplets will condense on the wafer’s surface. Tilting the wafer in VPD will cause these droplets to pool. Otherwise, an additional scanning solution can be used to help collect the droplets spread across the wafer’s surface. The silicon wafer surface, which becomes hydrophobic, will also facilitate the complete collection of droplets. After the collected droplets are heated and dry on the wafer’s surface, their contents can be analyzed. The composition of the scanning solution will vary slightly depending on what is needed. Typically, in addition to hydrofluoric acid, they will contain hydrogen peroxide to help improve Cu recovery [6]. This can effectively improve the signal during the VPD process. In addition, gaseous SiF₄ and volatile H₂SiF₆ will escape before measurement due to heating or because of their own volatility, which helps to reduce the interfering/inhibiting effects of Si matrix dissolution on the assay process.

The efficacy of VPD is mainly determined by two factors: the dissolution efficiency of the acid solution in regard to the contaminants and the recovery efficiency of the acid solution. Its concentration ratio can be easily assessed by measuring the surface area of the wafer relative to the area where the droplets dry. Since the surface contaminants originally distributed at different positions across the entire wafer are collected and concentrated into a single point for measurement, although the system’s sensitivity is improved, it loses the Spatial Resolution advantage of the original fixed-point analysis. Also, if the overall metal concentration is too high when the sample is concentrated, the Cu recovery will be less than 50%.

Considering the need for Spatial Resolution, instead of collecting the dispersed droplets with a scanning solution, the wafer can be heated directly so that the condensed droplets dry in place and are then measured directly. Although the results show that this method’s detection ability is not as good as that of the VPD system, the difference is only about 1.5–5.0 times, and this method retains the advantages of spatial resolution [3]. Studies which used SR as the light source showed that using higher intensity light sources leads to better detection limits. Also, by using radiation below the Si edge, it is possible to excite the Al atoms on the surface instead of the Si atoms on the substrate, enabling the specific detection of Al contaminations above 10¹⁰ atoms/cm² [7].

In Taiwan, there are only a few places with commissioned TXRF resources. The SR mentioned above is also not a widely installed light source device. SR light sources in Taiwan are mainly located at the National Synchron Radiation Research Center (NSRRC). This limits the convenience and practicability of using it to increase TXRF sensitivity.

In contrast to the rarer TXRF, ICP-MS has been widely used in environmental science, biology, forensic science, materials analysis and many other fields since it was developed in the 1980s. It destroys sample matrix components using high temperature plasma then analyzes it via a high resolution and high sensitivity mass spectrometer. When connected to different devices in series (hyphenation), it can fulfill many specific analysis and detection needs.

With the exception of halogens, whose poor ionization efficiency make this tool less applicable, ICP-MS can be used to detect low mass elements unsuited to or undetectable by TXRF, such as sodium, magnesium, aluminum, lithium, beryllium, and boron, etc. When detecting metal contamination on 8 and 12-inch silicon wafers, the detection limit for most metals generally falls within the range of 10⁸ to 10⁹ atoms/cm².

ICP-MS is commonly connected in series with the Laserablation (LA) system, Arc system, or Spark system to enable the direct measurement of solid (or biological) samples. If combined with Gas Chromatography (GC), it enables separation and measurement of organotin and organolead gaseous compounds. When used in combination with Electro Thermal Evaporation (ETV), in-line removal of sample matrix can be achieved via a Temperature Program and Matrix Modifier to complete the analytical testing of small samples. However, when not used in series with the abovementioned special equipment, ICP-MS requires samples to be converted/digested into a liquid form before analysis can be performed. Therefore, this is considered destructive analysis and cannot provide fixed point analysis information on samples.

In terms of the needs for wafer surface metal contamination detection, ICP-MS requires the native or deposited silicon oxide or silicon nitride film on the wafer’s surface to first be decomposed using hydrofluoric acid. The acid is then collected and analyzed, and the results are converted into the surface concentration unit of atoms/cm². The decomposition of the surface film layer can be done manually by dripping acid on the sample or with automated machine assistance (VPD). The manual method is also referred to by a variety of other names such as the rolling droplet method, droplet walkthrough, Direct Acid Droplet Decomposition (DADD), and Liquid Phase Decomposition-Droplet Collection (LPD-DC).

As for the VPD system, as described in the earlier VPD-TXRF paragraph, the saturated hydrofluoric acid vapor is introduced to react with the hydrophilic silicon oxide layer. The exposed hydrophobic wafer surface facilitates the scanning solution’s rolling and collection of surface droplets. The composition of the scanning solution will vary from experiment to experiment, but it will usually contain hydrogen peroxide, which can help improve Cu recovery. Excess hydrofluoric acid vapor and added oxidants, such as nitric acid or hydrogen peroxide, can promote the formation of volatile SiF₄ gas in the silicon oxide matrix, thus reducing the matrix interference and complex ion interference that may be caused by large amounts of Si matrix in ICP-MS measurements, such as ⁴⁷Ti(²⁸Si¹⁹F), ⁶⁸Zn(⁴⁰Ar²⁸Si), and ⁴⁴Ca(²⁸Si¹⁶O).

However, since the surface oxide layer treatment is done using gas, the thickness of the film will affect the length of the exposure time. Typically, native oxides (1.5–3.0nm) will take 20 to 30 minutes, while oxide layers of more than 10nm may require 3 to 12 hours of exposure. The manual pre-processing method can reduce the overall film etching time by simply increasing the amount of hydrofluoric acid used.

VPD systems typically come with robotic arms for processing samples. This helps prevent the possible introduction of contamination by human sample handling. The interior of the machine is usually a Class1 environment in terms of the level of cleanliness. Although the background value and detection limit of the assay can be effectively reduced, doing so makes the machine’s construction expensive. Therefore, though it is easier for additional contamination to be introduced when using the manual rolling droplet method, it is still widely used by laboratories because of its simplicity, low cost, speed and flexibility.

ICP-MS has been used in combination with the rolling droplet method or VPD for the analysis of metal contamination on wafer surfaces for many years. However, there are still some unresolved problems in regard to certain testing needs. Some of these issues include its applicability in the compositional analysis of non-silicon wafers, such as GaAs and GaN wafers, ceramic, sapphire, glass and quartz wafer carriers, samples of silicon wafers where the surface is not SiOx or SiNx, and thin films like Poly-Si, epitaxial silicon, SiW and Ti. Other issues to be considered include the possibility of optimizing the conditions required for the detection of precious metals (Pd, Au, Pt and Ru, etc.) and ICP-MS’s inapplicability when it comes to pattern wafers and wafer edge and bevel inspection requirements [8].

Currently, the samples used in silicon wafer surface metal contamination analysis are mostly silicon wafers with native or chemical grown and thermal grown silicon oxide or silicon nitride films on the surface. Such silicon wafer samples can be pre-treated by the aforementioned manual method or by VPD machine. Although thicker films require increased amounts of acid and reaction times, the hydrophobic nature of the film after etching can help determine whether the etching should be terminated.

For hydrophilic wafer surfaces, however, it is impossible to use this hydrophobic phenomenon to determine when to terminate film etching. What’s more, the substrate for some gallium arsenide and gallium nitride wafers will also dissolve slowly in acid systems containing hydrofluoric acid. Prolonging the pre-treatment time will thus result in an increase in the dissolution of the corresponding substrates. Although there is a limit to how much metal contamination can increase in high purity wafers, the dissolved substrate components can still affect the data collected by the detection equipment through both mass spectrometry interference and non-mass spectrometry interference effects.

Using a diluted acid as the extraction solvent can prevent the dissolution of the substrate, but it may be unable to effectively dissolve and remove contaminants from the wafer’s surface. The literature discussing the cleaning effects of different acid solutions on the surface of GaN wafers under different PH values and redox potentials claims that solvents with low PH and high redox potential can help reduce metal contamination on GaN surfaces [9].

These discussions are similar to those on the cleaning efficiency of Cu on wafer surfaces in other literature. In the future, the cleaning, removing and concentration-monitoring of metal elements on the surfaces of non-silicon wafers will help to further our understanding of the characteristics of wafer materials and their oxidation reduction potential. Studying the PH of the collected solution will contribute to achieving these goals. A certain degree of difficulty can be anticipated for the experiment design and validation in this area, including knowledge of sample background information and whether a laboratory has suitable testing equipment. Also, taking wafer recycling into consideration, it is not possible to use hydrofluoric acid solutions for sample processing in the few cases where ceramic, glass or sapphire, etc. are used as wafer carriers. Only diluted nitric acid solutions can be used for surface contamination collection.

Although you can consider oxidizing and decomposing films that cannot be dissolved using simple hydrofluoric acid, such as Poly-Si film, with mixed acids containing nitric acid and hydrofluoric acid, using higher concentrations of acid can cause violent reactions and lead to concerns of further etching into the underlying wafer. If the acid concentration is too low, however, it may not ablate sufficiently. As such, contamination detection of such films is difficult to perform manually. The VPD machine offers a bulk etching option that can simultaneously introduce ozone and hydrofluoric acid so that the sample’s composition can be analyzed while etching to meet the demand for detection of metal components in Poly-Si wafers. It has been pointed out in the literature, however, that it is easier for sample surfaces analyzed via the Bulk etching option to exhibit obvious roughness and hydrophilicity. This makes it easy for the scanning solution suspended at the end of the scanning nozzle to flow out [10].

Due to the use of and residue left by photo-resistors in the manufacturing process of silicon wafers with patterns, the acid used can become limited to the patterns. In addition to the possibility that the acid recovery rate will be low or even impossible, this makes it difficult for the acid solution to roll smoothly on the surfaces of these wafers, which also leads to differences in the etching time and depth at each position. However, there is currently no better solution for the inspection of such wafers. Soaking them in an acid bath is one possibility, but doing so would result in contributions from the wafer’s back side. Another possibility is to use a large amount of acid to cover the entire surface of the test piece, but this would mean a very high dissolution ratio, which would increase the possibility of Not Detectable (ND) results.

Due to the use of and residue left by photo-resistors in the manufacturing process of silicon wafers with patterns, the acid used can become limited to the patterns.

The two different sample tests above are difficult to conduct using either VPD-ICP-MS or LPD-DC-ICP-MS. We can consider evaluating the possibility of direct sample analysis via TXRF. There would still be the problem of a high detection limit, but the needs of the selection of acid components and evaluation of pre-treatment times could be avoided.

The main consideration in the analysis of certain analytes is the detection of precious metals. HF acid must be used for the pre-treatment of precious metal analytes such as Ir, Ru, Pd, Au and Pt to make the silicon wafer surface hydrophobic before the nitric acid and hydrofluoric acid are introduced. Use an acid similar to the Aqua Regia recipe for the scanning solution in order to raise the recovery rate of the analyte. However, note that the optimum concentration of hydrofluoric acid and nitric acid for different precious metals is different. The recovery rate generally falls between 20 to 80% [10-11]. For similar sampling techniques discussed in the reference materials, a recovery of 74 to 98% can be achieved in the detection of Au, Pd, Pt and Ag [12].

As for wafer Edge and Bevel measurement requirements, they are mainly used to evaluate the degree of contamination introduced by the areas where a wafer comes into contact with the wafer cassette. The accurate assessment of the distribution of contamination can increase the ratio of a wafer’s surface that can be used. However, without the help of special tools, ICP-MS has difficulty analyzing these two regions [12]. Depending on their needs, laboratories can consider carving a groove of a specific depth on the surface of a PFA material object to limit the acid to within a certain distance from the edge in order to sample certain locations. Even with the help of such self-designed tools for pre-processing, however, the absolute amount of analyte that can be obtained is still limited by the small sampling area. Therefore, when it comes to the contamination assessment of these two locations, it may be simpler and more direct to use TXRF for fixed point analysis [13].

In conclusion, metal contamination analysis of wafer surfaces can assist in the evaluation of the contamination of wafers caused by passing through a certain process, specific machine or even an entire process cycle. At present, the VPD-TXRF, VPD-ICP-MS and LPD-DC-ICP-MS used for the inspection of 8 and 12-inch wafers all have the potential to meet the detection requirement of 10⁸-10¹⁰ atoms/cm². For some elements, the detection can even reach 10⁷ atoms/cm² or lower.

Among these tools, TXRF is the one able to meet the need for fixed-point detection without the need for special sample preparation. Excluding those low mass elements to which it is inapplicable, its detection limit generally reaches the 10¹⁰ atoms/cm² level. Furthermore, since it is nondestructive to samples, it allows for repeated testing and verification of the same samples. Therefore, it is often used for monitoring production lines. If further combined with VPD for sample pre-treatment, though its original fixed-point detection ability will be lost, the detection limit can go all the way to 10⁸-10⁹ atoms/cm², which meets the current metal surface metal detection needs.

The more common ICP-MS machine, with its better detection capabilities, can be used to detect the low mass elements that are difficult to detect using TXRF. However, it cannot be used as a surface analysis tool for direct detection of wafer surface contamination because the silicon oxide or silicon nitride films on sample surfaces must be dissolved with an acid solution before testing can be done. Therefore, facing future wafer samples of other materials or special film samples, it is necessary to continue exploring the formula and applicability of acid solutions.

Reference

[1] Jennifer A.S, Lindsey H.H, (1995) TXRF Analysis of SC-1 Treated Silicon Wafers. J. Elecchem 142; 1238-1242.

[2] Yamagami M, Ikeshita A, Onizuka Y, Kojima S, Yamada T, (2003) Development of Vapor Phase Decomposition-Total-Reflection X-ray Fluorescence Spectrometer, Spectrochimica Acta Part B 58; 2079-2084.

[3] Takahara H, Mori Y, Shimazaki A, Gohshi Y, (2010) Method and Mechanism of Vapor Phase Treatment-Total Reflection X-ray Fluoroscence for Trace Element Analysis on Silicon Wafer Surface. Spectrochimica Acta Part B. 65; 1022-1028.

[4] Takaha H, Mori Y, Shibata H, Shimazaki A, Shabani M.B, Yamagami M, Yabumoto N, Nishihagi K, and Gohshi Y, (2013) Vapor Phase Treatment-Total Reflection X-rray Fluoroscence for Trace Elemental Analysis of Silicon Wafer Surface. Spectrochimica Acta Part B. 90; 72-86.

[5] Baur K, Brennan S, Pianetta P, Opila R, (2002) Looking at trace impurities on silicon wafers with synchrotron radiation. Analysis Chemistry 1:609A-616A.

[6] Danel A, Lardin D, Giroud C, Tardif F, (2003) Contribution for the Optimization of the Vapor Phase Decomposition Technique. Materials Science and Engineering B 102; 213-217.

[7] Laszlo F, Siegfried P, Ludwig K, Peter W, Christina S, (1999) Novel Method of TXRF Analysis for Silicon Wafer Surface Inspection. Fresenius J Anal Chem 363; 98-102.

[8] Fontaine H, lardine T. (2013) Quantitative Analysis of the Metallic Contamination on GaAs and InP Wafers by TXRF and ICPMS Techniques, ECS Transaction 58; 327-335.

[9] Nagao N, Nakamura K, Teramoto A, Shirai Y, Imaizumi F, Suwa T, Sugaw S, Ohmi T, (2015) Surface Metal Cleaning of GaN Surface Based on Redox Potential of Cleaning Solution. ECS Translocation 66; 11-21.

[10] Ichikawa M, Kishi Y, Ichinose T, Kawabata K, (2021) Analysis of Metallic Impurities in Si Wafers Using Fully Automated VPD-ICP-MS. PerkinElmer Application Note 1-6 .

[11] Devita M, Fontaine H, Drogue N, Mathiot D, Enyedi V, Lartin D, (2015) Collection Effficiency of Nobel Metallic Contaminants on Si Wafers HF-Aqua Regia Mixtures for VPD-DC ICPMS Analysis. Solid State Phnomena 219: 268-271.

[12] Scott A, Meredith B, (2010) Monitoring Wafer Cleanliness and Metal Contamination via VPD ICP-MS: Case Study for Next Generation Requirements. Microelectronic Engineering 87: 1701-1705.

[13] Hiroshi K, Motoyuki Y, Joseph F, Liyong S, (2009) Detection of Metal Contamination on Silicon Wafer Backside and Edge by New TXRF Methods. AIP Conference Proceedings 1173; 67-71.