Wafer manufacturers have been using mercury-probe or capacitance-to-voltage—Schottky (HgCV) techniques to measure the resistivity of wafers sampled from production runs. While these methods adequately measure a wafer's resistivity profile (resistivity as a function of depth into the silicon), any measured wafer must be scrapped, because of surface damage or metallic contamination, and manufacturers may lose as much as 4% of a production run for such sampling. To be qualified as a production-worthy alternative tool, however, the noncontact system must be able to demonstrate equal or improved measurement capability over a broad range of devices compared with current metrology techniques, without any wafer damage or contamination.

The project described in this article evaluated the Epimet Model 2 noncontact resistivity monitor from SemiTest (Billerica, MA) for its suitability to monitor epi wafer production. Designed for precision and accuracy, the system uses a column of air in place of the mercury used in traditional tools. Its failure modes are almost exclusively shutdown or a zero reading, which is advantageous because there is little or no measurement drift before failure. The dual focus of the research was to determine the measurement capability of this noncontact system and to answer the key question of whether the technique alters the sampled wafer in any way, causing it to be rejected.

Achieving reliable measurements with the noncontact system requires a surface preparation process analogous to what is required for sample measurement on HgCV tools. While the surface preparation technique for HgCV testing does not require that product quality be maintained, however, wafers prepared in the noncontact tool's surface pretreatment chamber (PTC) must meet this criterion.

In order to gauge the feasibility of the pretreatment and measurement protocol, a two-part evaluation plan was implemented. First, capability tests were conducted to determine the measurement performance of the tool. Stability, bias linearity, repeatability and reproducibility, and correlation to the performance levels of existing HgCV tools were all quantified and summarized.

Then a battery of tests were performed to determine whether any wafers measured by the noncontact resistivity tool had been damaged. Evaluations included analyses of gate-oxide integrity (GOI), measured by J*t testing (which measures current density per time in coulomb/cm²); metal contamination levels (MCL), measured by vapor-phase decomposition, inductively coupled plasma–mass spectrometry (VPD ICP-MS); minority carrier lifetime, measured by microwave photoconductive decay (µ-PCD); microroughness, measured by scanning force microscopy (SFM); and particles added, measured as light-point defects (LPDs) using a Surfscan 6200 from KLA-Tencor (San Jose). These tests cover potential chemical, mechanical, and electrical alteration to the silicon epi layer. The noncontact tool was designed to measure 100–200-mm wafers; 150-mm wafers were used for the tests discussed here.

Measurement Recipe Setup. When comparing the results from the noncontact tool with those from HgCV testing, there are several factors that need to be considered. Ideally, a direct comparison can be obtained by measuring the same points on the same wafer with both kinds of instrument. Therefore the measurement pattern across the wafer for the noncontact system was set up to be consistent with the HgCV probe pattern, including the same edge exclusion. Another factor that should remain constant is the method used for averaging the dopant concentration before converting results to resistivity data. For HgCV probes, an average is taken of a section of the dopant concentration over a specified depth region. Although the depth regions used in the measurement recipes in the comparison testing were not identical, they were representative of the flat profile segment in the epi-layer dopant concentration over a fixed depth.

Measurement Capability Test Protocol. The protocol for measurement capability testing was based on QS9000 guidelines for a variable measurement system study. Samples represented three resistivity ranges each for n- and p-type dopants, as indicated in Table I. The QS9000 standard states that measurement repeatability and reproducibility variation must take up no more than 30% of the specification range, with a target of 10%. This goal is challenging with HgCV techniques because of the variability inherent in those measurements.

Product Integrity Verification. Because HgCV test wafers are scrapped, no correlation testing was required in the area of postmeasurement product quality. However, since the goal of implementing noncontact monitoring is to measure actual product wafers, it was important to determine whether the sampled wafers are damaged by any step in the measurement protocol. In this study, the LPD, µ-PCD, and MCL analyses were performed by partitioning the wafer process flow so that deviations in wafer quality could be traced to individual process steps. A GOI test also was performed to verify that no other problems existed after wafer resistivity was measured by the noncontact tool.

Stability. The standard deviation method was used to determine the noncontact tool's stability when measuring n- and p-type dopants in various resistivity ranges. Table I lists the sample parameters. Three measurements were taken on each sample every 12 hours over a 25-day period. The QS9000 acceptance criterion is that the measurement-process ratio is <30% of the product specification; and the noncontact resistivity tool passed for all six samples. The results are summarized in Table II, with typical data plots shown in Figures 1 and 2.

Figure 1: Stability results for sample A: (a) x-bar data, and (b) range data.

Figure 2: Stability results for sample B: (a) x-bar data, and (b) range data.

Bias Linearity. Bias is the difference, or offset, between a measurement and a known value, while linearity is the fit of a regression line to different offsets over the tolerance of the tool. Ideally, the bias should be the same whether one measures resistivity at 1 or 60 W-cm. The QS9000 standard mandates that the regression line slope should be no greater than 0.03 and the fit of measured to known values should be greater than 0.95. Bias linearity is normally evaluated using an independently established standard (from the National Institute of Standards and Technology [NIST]); in this study, three p-type substrate standards from VLSI Standards (San Jose) were used—RS801, RS810, and RS 860—which have resistivity values of 0.861, 10.48, and 56.40 W-cm, respectively. The regression line was calculated and compared with these standard values, yielding QS9000-compliant results, as seen in Figure 3.

Figure 3: Bias linearity data for the noncontact resistivity tool measured against p-type VLSI standards of 0.861, 10.48, and 56.04 W-cm.

Repeatability and Reproducibility. The repeatability and reproducibility (R&R) of the tool measurements were assessed using the average-and-range method, which is often used to determine operator-to-operator variation. Because the noncontact system features automatic loading, the component of variation usually assigned to the operators instead represents a factor of drift or sensitivity to evolving environmental conditions (humidity, barometric pressure, and so forth). All six samples were assessed against typical customer specifications for epi wafers. The acceptance criterion is that the ratio of R&R to the specification tolerance must be less than 30%. The noncontact tool showed acceptable repeatability and reproducibility in all cases. These results are summarized in Table III.

Correlation. To assure that epi wafers manufactured to customer specifications are not affected by a change in metrology, it is critical to correlate the capabilities of the new instrument to those of existing instruments. For this purpose, measurements were taken on five sample wafers covering the full resistivity range over five days using both the noncontact tool's protocol and the HgCV protocol. In this comparison, a single-point calibration curve was established using the results for a high-resistivity wafer measured with the HgCV method as the standard or baseline.

Considering the large variability typically experienced in HgCV reading, this clearly is not the best means of calibrating a more-precise gauge. However, it is reasonable for purposes of comparing the two techniques. (Under production conditions, the noncontact tool is calibrated to NIST standards.) When the daily measurements were averaged and correlated, the resultant linear fit yielded a slope of 1.0155, with R² = 0.9958, as shown in Figure 4. The variation in the HgCV data (shown horizontally) was much broader than the variation in the noncontact tool data (shown vertically). This correlation meets SUMCO's criterion for technology transfer.

Figure 4: Correlation between noncontact-tool and HgCV measurements for n- and p-type samples. Correlation between the two gauges met in-house criterion for product and recipe transfer.

Product Integrity Verification Results

Samples for each product integrity test were taken from one run of material at four different points: directly after epitaxial deposition; after PTC processing; after PTC processing and measurement; and after PTC processing, measurement, and SC-1/SC-2 cleaning. The epi-only samples served as reference wafers.

Particle Performance. Light-point defects are a measure of the number and size of particles on a wafer surface. Figure 5 shows the average number of LPDs in six size ranges from 0.12 to 20 µm on the reference wafer and on samples subjected to various processes. LPD specifications vary greatly between customers; however, the data indicate that wafers conditioned with PTC or both pretreated and measured on the noncontact tool would be acceptable for most specifications. All samples would pass current international particle specifications for 150-mm wafers.

Figure 5: LPD test results for a reference wafer and samples from three process-flow points. The data for bins 1 through 6 represent increasingly large particles, with bin 1 being ≤0.12 µm and bin 6 ≤20 µm.

Carrier Lifetime. To determine if one or a combination of the process steps in the measurement protocol would cause carrier lifetime degradation, two wafers from each of the four sampling points were evaluated using µ-PCD. The data shown in Figure 6 suggest that none of the partitioned process steps are associated with minority carrier lifetime degradation. In fact, wafers subjected to some or all of the protocol exhibited better lifetimes (on average) than did the reference wafers and cleaned samples.

Figure 6: Results of µ-PCD measurements on two reference wafers and two wafers each from three sampling points. All samples showed acceptable minority carrier lifetimes.

Metal Contamination. VPD ICP-MS testing was performed to evaluate the surface metal contamination contributed by each step in the measurement protocol. As shown in Figure 7, measured surface concentrations for all elements were below 1.5 x 10¹⁰ atoms/cm². While specifications vary by customer and element, this threshold is considered acceptable by the industry.

Figure 7: VPD ICP-MS results for a reference wafer and samples from three process-flow points. Specifications vary by element; however, all of the metal contamination levels shown would be considered acceptable.

Electrical Effects and Surface Morphology. Effects of the noncontact tool's measurement protocol on gate-oxide integrity and surface roughness also were determined, using J*t measurements and SFM. The resulting data, shown in Figures 8 and 9, indicate that there is no difference between measured epi wafers and nonmeasured wafers.

Figure 8: Typical J*t yield-map data obtained from GOI testing. All pretreated, measured, and cleaned wafers tested showed acceptable yields, comparable to those of reference samples.

Figure 9: RMS roughness data measured using SFM on a 1 x 1-µm area on reference wafers and wafers measured using the noncontact tool's protocol.

Conclusion

The capability test results reported here indicate that resistivity measurements performed by the noncontact system are more repeatable and stable than those performed using the current HgCV protocol. The tool exhibited less variation in all measurement ranges explored in the study, which represent typical MOS epi wafer product specifications for n- and p-type dopants. The system also exceeded the performance and capability of the HgCV technique for resistivity measurements above 1.7 W-cm.

In addition, product integrity verification testing demonstrated that use of the tool was not associated with any contamination problems, and a gate-oxide integrity test indicated that it had no deleterious effect on device manufacturing yield. It should be noted, however, that product integrity is closely tied to the tool's maintenance history. Poor handling practices or incorrect operation can lead to hardware contamination, resulting in deviations from the performance levels captured in this study. Extreme care and rigorous requalification of the system is needed when maintenance activities breach the pristine wafer-handling areas of the PTC or the measurement chuck.

When the new metrology is implemented in ongoing, continuous manufacturing operations, reliability will become as great a concern as the system's capabilities and the maintenance of product quality. The tool's overall reliability and uptime measured by mean time to failure and mean time to repair will be assessed in the next phase of the work.

This article is based on a poster paper originally presented at the 13th Annual Advanced Semiconductor Manufacturing Conference (ASMC) on April 30, 2002, in Boston. The authors would like to thank Greg Martel, a technical representative from SemiTest who contributed his expertise on the Epimet noncontact resistivity monitor.

Christopher Panczyk, PhD, is the director of quality at SUMCO, where he is responsible for the management of quality engineering and quality assurance activities and programs for multiple manufacturing facilities on the West Coast. He is also the acting director of operations for the Fremont division, managing both the manufacturing and maintenance departments. Panczyk started his career in R&D at Sumitomo Sitix Silicon in 1996. He is a member of the Materials Research Society. He received a BS in chemical engineering from the University of Illinois at Urbana-Champaign and an MS and a PhD in chemical engineering from Purdue University in West Lafayette, IN. (Panczyk can be reached at 510/440-3334 or cpanczyk@sumcousa.com.)