Although the incidents of tool-part contamination have been identified as causes of device defects and high particle counts, ongoing correlations of process problems with parts contamination have not been possible because parts cleanliness is not normally quantified and reported. For early, large-geometry device processes, the impact of tool cleanliness variability was likely minimal, and the high process yields that were achieved delayed the development of quantification practices for parts cleanliness.

Over time, other reasons have also contributed to the delay. One significant factor has been the lack of appropriate methods for measuring ionic and organic contamination on tool surfaces. Baseline units for reporting surface cleanliness are dependent on the analytical methods used. X-ray photoelectron spectroscopy (XPS) reports in atom%, while acid extraction/inductively coupled plasma-mass spectroscopy (ICP-MS) reports in nanograms or atoms per square centimeter, based on an extract from the surface of the part. However, neither method is well suited for use with tool parts: the former can be destructive because of the required sample size, and the latter is itself
contaminating.

Another contributing factor to the delay in quantifying parts cleanliness has been the lack of institutional focus on the issue. Semiconductor tool OEMs typically subcontract the manufacturing of new parts to numerous machine shops, which do not follow the particle-free and ion-free manufacturing protocols found in the semiconductor industry. In fact, lubricating oil and particle-generating cutting processes are commonplace in machine-shop environments.

Attempts by OEMs to clean outsourced parts in-house prior to final shipments or to require machine shops to clean parts and then package them in Class 100 environments have no doubt been helpful, but data suggest that such parts do not always meet desired cleanliness baselines. In response, some major tool OEMs are beginning to insist that machine shops achieve stringent surface-contaminant specifications measured in terms of atomic percentage or in nanograms or atoms per square centimeter.

Additionally, until the recent emergence of the contract parts-cleaning industry, cleaning tool parts was done in subfabs as a necessary, but costly, noncore fab activity. Because device manufacturing received most of the available research funds, little was spent on developing ion- and particle-free parts-cleaning methods. As a result of all of these factors, tool parts in general missed the ultraclean revolution that affected semiconductor materials.

This article reports on a study that begins to address the issues involved in quantifying the cleanliness of tool parts. In the study's first phase, surface-measurement results from XPS, acid extraction/ ICP-MS, and DI-water extraction/ion chromatography (IC) were correlated, and the limitations and benefits of each method were reviewed. Next, "out-of-the-bag" parts were analyzed to establish an OEM quality-level baseline. Finally, tool and wafer contamination data were reviewed to evaluate the effects of utilizing quantified ultraclean parts in a sub-0.2-µm device etch process. Not surprisingly, the initial correlations indicated that cleaner parts improve process performance.

One factor limiting the quantification of tool-part surface cleanliness on a part-by-part basis has been the lack of suitable analytical techniques. XPS has been used for many years to identify and quantify the surface composition of materials. In this method, x-rays are used to excite the surface of a sample, and the emitted electrons are analyzed to obtain information about the elements present. Elements in the top 10–20 nm of the sample are detected efficiently. However, because it generally requires small sample sizes (20 x 20 cm), XPS is a destructive technique when it is used to quantify larger parts. Thus, the method is effective for quantifying and qualifying cleaning methods, but cannot be used directly to analyze the cleanliness of tool parts.

In recent years, surface extraction combined with ICP-MS has become a popular way to quantify contaminants because the instruments involved are relatively inexpensive and widely available. The technique uses a liquid extract obtained by washing the sample surface with dilute acid. The extract is then vaporized and ionized in a plasma, which is fed to a mass spectrometer for determination of each element's concentration. Because the entire sample is generally immersed in the dilute acid, the method is suitable for quantifying and qualifying particular cleaning methods, but is unsuitable for in-process analysis of tool parts because the parts must be recleaned after being immersed in the extraction solution.

A third technique, DI-water extraction/IC, can be used to test surface ionic cleanliness using only pure water as the extraction solvent. The aqueous extract of the sample surface is concentrated and injected into a small ion-exchange column, where the elements are separated. Detection is based on the conductivity of the ions, and the method is capable of detecting both acid residues (e.g., fluoride and chloride) and alkali metals (e.g., sodium and calcium). An advantage of DI-water extraction/IC is that sampled process-tool parts need not be recleaned, because the DI water is not in itself contaminating.

All three methods offer insights into surface cleanliness, but each has inherent limitations for verifying the cleanliness of tool-part surfaces. None of the three can detect particles, a critical measure of parts cleanliness. XPS can differentiate between the chemical states of the atoms (e.g., inorganic versus organic fluorine) but cannot identify the organic species. Table I summarizes the advantages and disadvantages of the three methods.

Sample Preparation and Experimental Results. To obtain real data from samples analyzed using these three techniques, aluminum coupons were cleaned simultaneously at QuantumClean's Advanced Cleaning Technology Center in Irving, TX, utilizing two different cleaning methods: an alkaline treatment and an acid/oxidizer treatment. Care was taken to ensure that all samples were cleaned, handled, and bagged in precisely the same manner. Duplicate coupons were sent to three laboratories, each with the capability to perform one of the three analytical techniques.

For XPS measurements, no further sample preparation was necessary. For ICP-MS, each coupon was leached with dilute nitric acid for a specific extended time period, followed by leachate analysis. For DI-water extraction/IC, the samples were wiped with a piece of polyester cleanroom wipe that had been prepped with DI water, and DI water was then used to extract ions from the wipe. The resulting acid extraction/ICP-MS and DI-water extraction/IC data were corrected for blank values to quantify the contaminant contribution from the target samples.

The elements that were measured with each technique are presented in Table II. As seen in the table, the XPS results were converted from atomic percent to atoms per square centimeter. This conversion process is not straightforward and is based on a number of assumptions. The x-rays that are aimed at a sample excite the atoms near the surface more efficiently than they do those below the surface, with the depth of analysis being dependent on the angle of the x-ray beam. For the purposes of this study, the response of atoms down to 10 nm was assumed to be good and the contribution from deeper analytes was assumed to be insignificant. Thus, the conversion results in the table are reported as if all the contaminants were on the surface. If the impurities were more evenly distributed throughout the XPS detection volume, then the true surface concentration would be an order of magnitude less than that reported here. The factor applied in the conversion is based on the density of aluminum and thus would be different for other substrates.

For the alkaline-cleaned aluminum coupons, there was a good match between the ICP-MS and XPS results for those metals that both methods detected, particularly calcium and magnesium. The lower titanium results from ICP-MS may have been a result of the insolubility of TiO₂. ICP-MS was able to detect sodium below the ~0.1 atom% limit of XPS. For the acid-cleaned aluminum coupons, ICP-MS found much higher contaminant levels than XPS; the reason for this difference is unknown. In most cases, the metals detected only by ICP-MS were found in quantities near or below the detection limit of ~6 x 10¹³ atoms/cm² of the XPS instrument. Only XPS was able to quantify the presence of carbon in the samples, which can indicate organic contamination.

XPS and ICP-MS detect base-material atoms in different ways. XPS collects a signal from the aluminum substrate itself, while ICP-MS detects only those aluminum atoms that are extracted during sample preparation. The higher aluminum levels seen by ICP-MS for both the acid- and alkaline-cleaned coupons suggests that some dissolution of the substrate did occur during the preparation process.

The data for elements measured by IC and one or more of the other techniques also can be compared. The DI-water extracts analyzed by ion chromatography showed less calcium and potassium than did the acid extract prepared for ICP-MS. Sodium ions are very soluble in DI water, and the levels found by IC and ICP-MS were consistent at the respective instruments' detection limits. The fluorine ion levels determined by IC were several orders of magnitude lower than those detected by XPS. That result could indicate any of several things: that fluorine was tightly bound to the aluminum/oxide surface, that it was present as an organic species, or that it was present below the surface.

Ion chromatography is the only technique of the three that can easily measure chloride ion contamination, which is commonly caused by improper sample handling and also can stem from acid cleaning-recipe residues. In these aluminum samples, chloride levels were <3 x 10¹², showing that control over inadvertent laboratory contamination had been good. Likewise, levels of nitrate (potentially from nitric acid) and sulfate (potentially from sulfuric acid) were below the detection limits of <2 x 10¹² and <1 x 10¹² atoms/cm², respectively. The higher levels of these elements found by XPS may not represent ions; their source cannot be determined from the data.

Discussion and Preliminary Conclusions. Although the sensitivities of the three analytical techniques are approximately within an order of magnitude of one another, none of the techniques can be applied universally. DI-water extraction/IC is the most benign technique, being nondestructive, noncontaminating, and compatible with even very large parts. However, if oxide levels are critical, only XPS can provide that data. For low levels of transition metals, ICP-MS is preferable. For acid residues and the contaminants associated with routine handling, only IC can easily detect the relevant ions. In addition, none of the methods can detect or speciate organic contamination or particles. Although correlations between results obtained using two or three techniques were seen in this study, it is best to compare before-and-after or between-part data derived using the same technique.

In order to make part-to-part comparisons, the study results were all expressed as atoms per square centimeter, a metric that enables the comparison of contamination on different substrates, such as aluminum, quartz, ceramic, and stainless steel, and is familiar to engineers in the semiconductor industry. Sample preparation methods were also standardized to be applicable to all substrates.

To baseline the variability of out-of-the-bag OEM parts, nine parts manufactured and cleaned by a variety of suppliers for one well-known toolmaker were unpacked and analyzed in a Class 10 cleanroom. All of the parts had been packed in multiple bags, which generally had stickers on them with such notifications as "Cleaned to High-Purity Specifications" or "Open in Cleanroom Only." To increase accuracy, two sample extracts taken from each of the nine parts were analyzed using DI-water extraction/IC. Four parts were also analyzed for particle contamination using a Dryden Engineering QIII aerosol surface particle detector (Pentagon Technologies, Livermore, CA).

A brief review of the results of the ionic tests, which are presented in Table III, indicates that parts cleanliness varied widely. In some cases, the variation between the low and high results for an element exceeded several orders of magnitude. For sulfates, for example, there was a 1900% increase from the lowest reported value to the highest. In addition, the absolute magnitude of some ions—such as sodium, which was detected at a level of nearly 10 x 10¹⁴ atoms/cm²—represents relatively high levels of those contaminants.

The test results from the four parts are shown in Table IV. It should be noted that some particulation may have been created by the friction of each part with its inner bag during the unpacking process. However, it is likely that with such a wide range of particle counts measured, the variability exhibited was statistically significant. Aerosol surface particle counting, while a good qualitative and semiquantitative particle measurement technique, is not absolute. Submicron particles can adhere to surfaces via various mechanisms, including electrostatic attraction, and thus cannot be counted using low-flow aerosol techniques.

After these out-of-the-bag analyses had been performed, parts were cleaned using high-purity cleaning techniques and reanalyzed. The IC results are summarized in Table V and the particle results in Table VI. For ionic contaminants, there was a significant decrease both in terms of absolute ion levels and in variability. (The differences in detection limits between the high and low values arose from using different-sized surface areas during sample extraction.) There also was a significant drop in particles measured on the cleaned parts.

Ensuring parts cleanliness is crucial for success in the semiconductor industry. Particulate, ionic, and human-related contaminants are among the largest contributing factors to decreased wafer yields and tool performance. Some tool operators indicate that fingerprints can outgas for hours while a tool reaches high-vacuum base pressures. In some cases, particles as small as 0.1 µm can cause device failure. Mobile ions such as sodium and potassium, metal ions such as iron and copper, hydrocarbons, and particles can easily be introduced during parts-manufacturing and -cleaning processes. Sources of such contaminants can include parts substrates, the machining process (released particles, cutting fluids, degreasers, etc.), the surrounding environment, human contact, the chemicals and water used to clean parts, postclean drying ovens, and the bags used to package parts.

The link between contamination and process tool performance is well understood for some semiconductor processes. For example, ionic and particulate contaminants on parts have been identified as the root causes of failure in diffusion processes. Typical analyses performed on wafers following tube/boat replacements in oxide processes have measured atoms per square centimeter and atoms per cubic centimeter levels of iron as well as mobile sodium and potassium ions, which can cause gate-oxide performance degradation. Additionally, total reflection x-ray fluorescence analysis performed on deposited poly/ nitride films following tube /liner/component changes has proven a reliable way to detect ionic contamination on new tool parts prior to production runs. Contaminant levels in the single-digit nanogram-per-square-centimeter range and atom levels in the ≤10 x 10¹² atoms/cm² range are required to achieve proper film performance. High-temperature oxide processes may require iron levels <10 x 10¹⁰ atoms/cm².

Tool performance in other semiconductor processes also can be correlated to parts cleanliness. For example, parts with high metal ionic contamination and/or high particle counts can prevent etch tools from performing properly, leading to low etch rates, poor uniformity, and end-point detection problems.

Figure 1: Micrographs of gas holes in an etch tool’s upper electrode. The improperly cleaned one (a) contains residual etch by-products, while the properly cleaned one (b) is free of contamination.

Achieving acceptable etch process results after a chamber strip can be difficult. Typically, cleaned chambers must first pass a mechanical particle-count qualification and then an RF/gas-on particle-count qualification. If the measured particulate levels are high, the chamber will have difficulty running to specifications. In a good wet-strip recovery, single-digit particle counts and low ion levels are possible if care is taken while cleaning the chamber and the newly installed parts.

Figure 2: Results of extraction/ICP-MS and particle tests performed on the electrodes shown in Figure 1.

Figure 3: Results of particle count tests performed on wafers processed after the electrode with the improperly cleaned gas hole shown in Figure 1a was installed in the etch tool (left side of chart) and after the one with the properly cleaned gas hole shown in Figure 1b was installed (right side of chart).

Conversely, incomplete cleaning can lead to contaminant levels that affect chamber performance. Figure 1a shows a gas hole of an upper electrode in which improper cleaning resulted in the incomplete removal of etch by-products, while Figure 1b shows a properly cleaned gas hole. The ion and particle cleanliness levels of these two electrodes are presented in Figure 2. Ionic contamination was measured using extraction/ICP-MS, while >0.3-µm particles were counted using the QIII instrument. Elevated levels of aluminum, calcium, iron, and fluorine, and relatively high counts of 40 particles/sq in., were detected on the improperly cleaned part. The aluminum/fluorine combination may indicate the presence of aluminum fluoride, an etch by-product, or a residue from the cleaning process.

The effect of these electrode parts on process performance can be seen in the >2.0-µm particle-count results for wafers shown in Figure 3. Even after a perfect wet-strip recovery was performed, the existence of the improperly cleaned electrode led to a process particle-count failure (large particles going out of specification) after the chamber had run 10 RF hours. This failure was likely caused by etch by-products depositing on the unremoved by-products in the gas hole and the resulting larger particles being released into the process chamber. When the properly cleaned electrode was installed, however, the large-particle counts remained within or close to the upper size limit (USL) for the full process run.

Tool parts are not routinely delivered with certificates of analysis verifying surface cleanliness, as is mandatory for silicon, chemicals, and gases. However, the combination of increasing wafer sizes, decreasing linewidths, and potentially contaminating materials such as copper makes it essential to quantify and report the cleanliness of tool parts. Only when part cleanliness is quantified can its impact on process performance be determined.

In terms of metallic ions and particles, surface cleanliness varies widely from part to part, even on components supplied by reputable OEMs and contract parts cleaners. While this article is by no means an exhaustive discussion of quantified part cleanliness, analytical methods, and the effects of part cleanliness on wafer processes, the correlations reported here support the intuitive belief that cleaner parts will lead to better process tool performance.

The authors wish to thank Ed Francis and Wayne Howell of National Semiconductor, Tim Burrows of Air Products and Chemicals, Dwight Zuck of QuantumClean, Jeff Stull of Desert Data, and Jeff Gold of Evans East for their invaluable assistance with this project.