Exhaust gas monitoring: New window into semiconductor processing

05/01/1999

Most semiconductor processes employ gases that induce chemical reactions with the semiconductor wafer to deposit or etch material, or that clean the processing chamber. Knowledge of the gas composition within the processing chamber or in the exhaust can provide important information about what is occurring during a processing or cleaning step. Optical emission spectroscopy (OES) has been utilized as a tool to measure in situ gas compositions within the processing chamber, while mass spectroscopy, residual gas analysis (RGA), and Fourier transform infrared (FTIR) spectrometry have been used in the exhaust duct.

The application of these analysis methods can provide information on the process behavior, etch end points, clean end points, plasma behavior, contamination levels, first-wafer effects, and chamber wall effects. This information is useful at all stages of the process life cycle, including tool development, process development, equipment installation or qualification, process tuning, chamber matching, routine maintenance, troubleshooting, process control, and fault detection and classification.

We will describe IR absorption spectroscopy, both FTIR and filter-based methods, and discuss their application to all stages of semiconductor processing. FTIR spectrometry is particularly useful in pre-production applications (e.g., development, tuning, troubleshooting, etc.), where knowledge of multiple gas species is important. FTIR multigas analyzers are fast, reliable, and
accurate and have constant calibration. They can provide compositions for complex mixtures at concentrations from sub-mtorr (ppm) to hundreds of torr, with measurement times under 1 sec.

Historically, it has been considered too costly to use FTIR for production monitoring of multiple gas species. Also, most production applications require monitoring of only one, or at most a few species. In these cases, FTIR can be employed during process development to determine the appropriate absorption bands for the analyte species and reference windows where no absorption occurs. A low-cost, filter-based, nondispersive infrared (NDIR) sensor can then be implemented for production-scale process monitoring and control.

Sensors based on IR absorption have advantages compared to OES and RGA sensors. OES sensors are difficult to calibrate because plasma emissions are sensitive to plasma power, and in some cases are nonlinear. In addition, OES sensors detect light generated within the plasma. This light is gathered through a window that is exposed to energetic, reactive atoms and molecules. Window coating and roughening can significantly degrade transmission, and hence the calibration and the sensor signal-to-noise (S/N) ratio. These effects result in drift over the sensor's life and the preventative maintenance cycle.

Typically, a significant over-etch margin is allowed to compensate for this sensor drift, lowering throughput and increasing chamber wear. Remote plasmas and thermal processes produce no light, however, and time and power settings are fixed based on laborious process testing. This approach severely penalizes the tool throughput, as the process times must be longer than the worst process case normally encountered. The IR sensors determine composition by light absorption rather than emission. Thus, IR sensors can be used for remote plasmas and thermal processes that cannot be monitored by OES. In this case, window deterioration will affect the S/N ratio, but not the calibration.

IR sensors have advantages when compared to RGA sensors as well. For example, they operate over a wider range of pressures, eliminating the need for expensive differential pumping systems. In addition, RGA sensors are more difficult to calibrate, the calibration isn't constant, and they require more maintenance than IR sensors. One reason for this is that both the RGA and the plasma itself are sources of molecular cracking that vary with the process conditions and sensor settings. This presents a fundamental problem in the quantitative interpretation of the RGA results. Another problem is that the corrosive gases that come in contact with the RGA's critical parts can cause damage to the sensor and result in calibration drift over time.

In comparison, the IR molecular spectrum is effectively constant, so spectrometer calibration for the analyte gases need only be done once. Modern FTIR spectrometers are available that allow transfer of calibrations between instruments. This eliminates the need for field calibration and for individual calibration of each instrument, which is a significant advantage. Reference spectra obtained in the absence of absorbing species can be measured periodically to account for detector and source intensity changes or deterioration of optical components. But since an IR spectrum typically contains regions where there are no absorbing species, each spectrum can provide a signal intensity reference.

The disadvantages of IR spectrometers are that they do not measure symmetric molecules such as N₂, O₂, and H₂, or single-atom species. In addition, the complexity of the IR spectrum historically required a trained user to interpret and quantify the species present. Automated spectral analysis software now available provides the interpretation and quantification of the complex data set required for complete on-line analysis.

Technology

FTIR spectrometry. FTIR spectrometry has been shown to be an ideal technique for quantitative analysis of complex gas mixtures from percent (torr) to sub-ppm (sub-mliter) levels. FTIR spectrometers have a fundamental advantage of larger optical throughputs when compared to dispersive spectrometers, and can obtain the entire spectrum in <1 sec. These features allow for rapid, real-time analysis of most semiconductor gas compositions with high sensitivity, even in complex mixtures. One other significant advantage inherent to FTIRs is their frequency accuracy. The data are collected with the aid of a single-mode laser used as a spectral reference. This allows data to be collected day after day and analyzed from multiple instruments using common libraries of reference gases with no change in frequency and no need for recalibration.

Within the semiconductor industry, FTIR spectrometry has been successfully used to monitor perfluorocarbon (PFC) process emissions during in situ plasma cleaning of a plasma-enhanced chemical vapor deposition tool [1, 2]. Multicomponent gas mixtures of CF₄, C₂F₆, C₃F₈, SiF₄, COF₂, and HF were simultaneously identified and quantified using extractive FTIR spectrometry. A single FTIR instrument is capable of performing quantitative analysis of etchants, plasma products, etch products, and contaminants. FTIR spectrometry has also been employed for stack monitoring, for testing the efficiency of abatement devices, for process development, and for the development of production filter-based NDIR sensors.

In a recent study, FTIR spectrometry was used to detect faults such as wafer contamination, air leaks, or incorrect wafers [3] by comparing the histories of the multiple exhaust gas species to those in standard runs to detect faults. For this study, FTIR gas measurements were made using an On-Line INDUCT system based on the Model 2100 process FTIR [4]. The spectrometer was positioned above the exhaust duct of the etch chamber between the turbo and mechanical pumps. The IR beam was directed into a multipass mirror assembly enclosed in an exhaust line cross. The optical arrangement is shown in Fig. 1. Spectra were collected at 1 cm-1 resolution and sixteen scans were co-added over 5 sec to provide approximately 11 data points/min. A liquid N₂-cooled mercury-cadmium-telu ride (MCT) detector was used. Fig. 2 presents a typical spectrum obtained during a standard CF₄ etch. Peaks for SiF₄, CF₃H, CF₄, COF₂, CO, H₂O, and CO₂ are identified. The peak-to-peak noise level across the fingerprint region was <0.002 absorbance units, allowing concentrations down to the 0.005ppm or 4µtorr level to be measured.

Quantitative analysis was performed using a classical least squares (CLS) analysis routine that was developed by On-Line. The routine uses selected spectral regions to minimize interference and to compensate for species with non-Beer's Law behavior. The CLS analysis compares the measured spectra to a library of reference spectra obtained for pure gases at known concentrations. The analysis of several species done during a series of wafer etches and chamber cleans (Fig. 3) showed time evolution of several exhaust gases in a Lam etcher employing HBr/Cl₂ chemistry. A series of bare Si test wafers and poly-Si etch rate wafers were run under standard etch conditions. Both the total concentration of etch products and the ratio of products changed with the number of wafers processed, with the largest changes occurring during and after the first wafer. These data demonstrated both a first-wafer effect as well as slow chamber-conditioning behavior.

NDIR. While some FTIR systems are too costly for production process control applications, filter-based, nondispersive techniques enable the manufacturing of low-cost, robust sensors that have the advantages of IR analysis and are ideal for the production environment. Although single-wavelength detection is feasible, dual wavelength sensors are preferred to permit real-time corrections for sensor and optical path changes. This typically can't be done with OES-based sensors. Optical emission peaks typically ride on top of significant broadband emissions, making determination of the sensor transfer function difficult or impossible. For the NDIR-based approach, spectral regions with no interfering absorption were found for every process studied to date, allowing for accurate determination of the sensor transfer function, and hence, nearly ideal signal normalization.

Response times for these types of sensors are typically on the order of tens of msec. Noise levels are in the microtorr range, with exact values depending on the specific design. For example, On-Line's Process Sense for SiF₄ has 50msec cycle time, sensitivities of 76 or 3.8mtorr/V depending on the output setting, and noise levels <10mV on both output settings. Multiple filters can be employed, allowing the analysis of several gases in the same sensor. Sensor prices are comparable to those for low-end RGA and OES sensors. From a service point of view, all the NDIR parts are outside the processing chamber or ducts, so sensor failures can be serviced without breaking vacuum seals. While window coatings can affect S/N, the calibration remains constant over the life of the sensor. Heated ducting or purged windows can be employed to reduce or eliminate window effects.

Relating processes to exhaust gas composition

Here we examine the relationship between exhaust composition and process-related variables such as products from plasma decomposition; products from etch processes; effects of wall seasoning; etch end-point detection; and chamber clean end-point detection. The examples come from studies in which FTIR was employed to monitor the various processeses; occurring during wafer etch. The experiments were performed in a Lam Research Corp. Research and Development Test Bench (RDTB) [3] and on a Lam etcher [5]. The RDTB is a large "high-flow" and "high-density" reaction chamber with good diagnostic access, which provides a flexible, high-performance test bench for reactor testing. PFC etch experiments were performed with the RDTB, while HBr/Cl2 chemistries and oxygen cleans were studied on the Lam etcher used for 150mm wafers.

Analysis of plasma products. The FTIR exhaust analyzer provides a detailed window into the chemistry in a plasma wafer etch or plasma chamber clean. As an example, a CF4 plasma etch illustrated in Fig. 2 resulted in a rich spectrum of exhaust compounds directly indicative of the chemistry occurring in the reactor. In that study [3], settings and conditions were varied to explore their effects on the plasma chemistry. To observe the effect of plasma properties, plasma power was varied during the etch of a bare silicon wafer. The SiF4 partial pressure (a measure of the etch rate) was observed as the plasma power was modified, and, as expected, the etch rate varied systematically with the power. It was also observed that a high plasma power minimized the formation of other PFC etch products, such as C2F6, which should be minimized for maximum CF4 utilization and environmental concerns. Thus, the FTIR provides a convenient method of optimizing the plasma power settings to maximize the active etch constituents.

Analysis of etch chemistry. FTIR analysis of exhaust gas composition is also useful in studying the processes occurring during the etch of a complex, multiple-layer patterned wafer. Cl2/HBr etch chemistry was investigated. In this chemistry, reactive Cl and Br radicals are generated to etch the Si (or SiO2) by forming volatile SiClxBr(4-x) species. The ratio of chlorine to bromine impacts both the etch rate and the etch selectivity for Si and SiO2. The higher the chlorine concentration, the more aggressive and the less selective is the etch. The process gas and etch product partial pressures were measured as a function of time during a 10-wafer sequence (Fig. 3). SiBr4 and SiCl4 were identified, and three mixed chlorobromides, SiClBr3, SiCl2Br2, and SiCl3Br, were also tracked. For this etch sequence, the 10 wafers were etched, starting from a cold standby mode. All the wafers except two were bare silicon. (The two were poly on oxide etch rate test wafers, and showed slightly different etch rates than the bare Si.) The data showed both a prominent first-wafer effect, as well as a longer-term drift attributed to chamber warm-up. Neither of the first two etchants reached steady-state values over the 10-wafer run. These results illustrated the sensitivity of IR exhaust gas analysis as a diagnostic tool for characterizing the complex and subtle chemical dynamics encountered in typical production.

In a second study, the partial pressures for SiF4 and a second compound C2F6 were measured [3] during a CF4 wafer etch that has 98% photoresist coverage over a 2.4?m oxide layer. The data showed three distinct regions separated by rapid changes in the two gas compositions. These rapid changes are the end points of the etch sequences. To interpret the data, gas compositions were determined for each of the surfaces that were etched during the process: the photoresist, the 2.4?m thermal oxide, and the bare silicon. The relative compositions of products in calibration runs on uniform samples with a single layer (or no layer) were compared to those at different times in the etch of the complex sample. Based on these comparisons, the etch sequence could be divided into three regions: Region 1, photoresist etch; Region 2, SiO2 etch; and Region 3, bare silicon etch.

A more complicated etch sequence was similarly interpreted. The wafer had a uniform 1?m-thick layer of SiO2 covered by photo resist over 65% of the surface. The concentrations of two compounds, C2F6 and CF3H present during the etch, showed distinct changes in concentration as the etch proceeded through the SiO2 and photoresist, and these variations were interpreted as follows. In Region 1, both the photoresist and SiO2 are being etched. Region 1 ends when the photoresist is completely removed. The etch during Region 2 is on SiO2, which is now exposed over the whole wafer. Near the end of Region 2, the area not originally covered by photoresist is now bare silicon. Region 2 ends when the original 35% of open area has been etched down to the bare silicon. During Region 3, the remaining SiO2 is removed, ending only when the wafer is bare. Region 4 is characteristic of a bare silicon surface etch. Such studies thus provide a convenient method to determine etch rate and selectivities.

Analysis of etch end point. One important monitoring need in etch processing is end-point detection. This is typically done using OES, a widely applied technique to measure the changing intensity of a chemically specific emission line. Although OES is a mature production-proven technique, there are several types of situations where its techniques are problematic or fail completely: 1) remote plasma and thermal processes that create no measurable light; 2) low etch area end point (low S/N with OES); and 3) trench etch processes that have no identifiable OES end point. Thanks to its ability for quantitative speciation, FTIR spectrometry is a superior alternative to OES in these areas of difficulty.

To illustrate the application of FTIR to low open-area etch, consider the CF₄ wafer etch discussed above [3], which had an initial open area of 2%. The effect of the open area can be seen in the product profiles in Region 2. Since the oxide was relatively thick (2.4µm), all the photoresist was etched before the oxide could clear. A rise in C₂F₆ in the middle of Region 2 can be identified, which corresponds to oxide clearing in the 2% open area not originally covered by photoresist, demonstrating the sensitivity of absorption spectrometry to low, open-area applications. In this same situation, traditional single-wavelength OES has difficulty in resolving low, open-area end point because the small signal from the chemical species of interest is masked by the high background from the other products and broadband background.

Analysis of chamber wall conditioning. The effect of chamber wall chemistry was investigated by varying the Cl₂ flow rate during a Cl₂/HBr etch of a bare Si wafer (Fig. 4). The run started with the standard etch rate test Cl₂/HBr process chemistry with flows of 20sccm Cl₂ and 100sccm HBr. During the run, the Cl₂ flow was modified. The first change doubled the Cl₂ flow to 40sccm. When the Cl flow was stopped and restarted, it took several minutes for equilibrium to be established. The persistence of product B (a mixture of chlorine-containing, Si-based compounds) indicates that the coating on the walls contributes chlorine to the plasma as the wall chemistry achieves a new chlorine-free steady state. It is also interesting that the final steady-state values for all three etch products differ from the original values. The results demonstrate the value of exhaust gas analysis in characterizing the impact of the process tool condition on process stability. Data of this type can be valuable when correlating yield with process tool variables.

Analysis of chamber cleaning. During deposition, materials are deposited in a layer on the chamber walls. If this layer is allowed to grow too thick, flaking occurs, depositing yield-killing particles on the wafers. For this reason, the chamber is periodically cleaned with a reactive, typically fluorinated, plasma etch. As the fluorinated ions and radicals impact the coating built up on the chamber walls, reactions proceed quickly and large amounts of SiF₄ are created. As the cleaning proceeds and nears completion, the SiF₄ concentration drops. The Process Sense NDIR sensor provides a signal proportional to the SiF₄ concentration in the exhaust stream. Unlike OES sensor approaches, the NDIR technology is applicable even when the clean process employs a remote plasma, because the sensor does not require emission from the analyte. The SiF₄ signal can be used for end point, turning the plasma off when the signal decays to a certain level. The active sensor-based end pointing can provide significant improvements in tool throughputs by eliminating timed overcleans.

First-wafer effect and chamber recovery. This phenomenon occurs when a new lot is run. The first wafer is usually significantly off spec as the chamber is "seasoned" to stabilize itself and the process. The name "first-wafer effect" is misleading, as it can apply to many wafers. For some processes, three or more dummy wafers are used to season the chamber before it processes product, incurring a significant chamber throughput penalty. Time traces like those in Fig. 3 illustrate the effect clearly. The analysis allows rapid quantification of the severity of the effect, and using exhaust gas analysis, the number of dummy wafers required to place the chamber into steady state can quickly be determined, resulting in reduced process development times and improved throughput.

Timed process step characterization. Using exhaust gas analysis, the times and flows required for pumping and flushing the gas delivery manifold can quickly be determined as well, leading to maximum throughput with minimum use of consumables. For timed remote plasma or thermal applications, one can easily determine optimal process times. In any of these cases, process variability can easily be measured, and optimized process timing can provide significant cost reductions through reduced consumable usage and increased tool throughput.

Applications of exhaust gas monitoring

Use of FTIR in R&D. The examples above demonstrate the range of phenomena that can be explored with FTIR spectrometry. The applications are straightforward for the development of etch and deposition processes, chamber cleans, end-point sensors, abatement requirements, plasma settings, and the development of fault detection and classification methods. Spectrometers can be installed on a cart for transportable extractive analysis. This configuration allows a single spectrometer to be used for many applications. One disadvantage is that the gas sample must be extracted into a gas cell, which increases the system's response time to minutes. When time resolution is important, the spectrometer needs to be set up for in situ monitoring as illustrated in Fig 1.

The IR spectral data provide information and answers to many important process development questions. What are the acceptable processing conditions? How can the window be maximized? What is the latitude for over-/under-etch and over-/under-clean? What is the required length of a chamber season? How can this be minimized? How often is cleaning required? How often is PM required? How severe are first-wafer and warm-up effects? FTIR multigas analyzers can provide quick answers to these and many other questions.

Use of FTIR during the equipment cycle. Once the process behavior is characterized by FTIR, the composition profiles for etch, deposition, chamber conditioning, steady state, cleaning, etc., are known. These can be employed as standards during the installation and qualification of the equipment, and during routine troubleshooting and maintenance. The gas composition profiles will ensure that the equipment is free of contaminating species or air leaks. The profiles will validate the sequence of deposition or etch is following the standard and that end points are reached on time. The data can be employed for chamber matching and post PM tuning and qualification, and will also reduce the time for installation, troubleshooting, and maintenance functions.

NDIR spectroscopy in production While the use of the FTIR for permanent production monitoring is often considered too expensive, many applications only require one or a few species to be monitored. FTIR can be employed to identify the appropriate spectral absorption bands for the species of interest and for a reference window where no absorption occurs. Then a low-cost, filter-based, NDIR sensor can be implemented for production.

As an example, consider silicon etch or clean using a fluorine etchant. In this case, SiF₄ is the dominant etch product. A version of an NDIR system (called Process Sense) was developed by On-Line specifically for SiF₄. Figure 5 shows the decay of the NDIR signal in an exhaust duct as the active product is pumped out of a test chamber in the laboratory. In production applications, the sensor is permanently mounted in an exhaust duct, while the signal is fed to the equipment's control system.

NDIR sensors can be configured for multiple species as well. This eliminates the need for multiple sensors on chambers equipped to run multiple process chemistries. These multigas sensors can be used for fault detection as well.

Fault detection and classification (FDC). This term applies to a broad class of techniques used to detect problems or faults in a specific process or group of processes [6]. In current fault detection systems, a training set of machine-state data (flows, pressures, temperatures, etc.) and process-state data (optical emission, plasma power, etc.) are collected during normal operation. The data are collected from the process tool either through the SECS port or from auxiliary sensors. The same machine-state and process-state data are collected in subsequent production runs. The FDC system then analyzes these data to determine departures from normal operation, and hence detect faults. If the process is deemed out of control, a message is sent to the factory control system to halt processing and display corrective actions for the maintenance personnel. While such systems have proved valuable, the range of acceptable data in normal operation can vary due to changes in fabrication tool components, shifts in tool properties resulting from use (deposits, parts wear, etc.) or overhaul, or variations in the optical emission window. This can result in false alarms, or undetected faults.

Because FTIR and NDIR spectrometry exhaust gas monitoring provides such a rich slate of information on factors such as contamination, process timing, process chemistry, end points, chamber-cleaning behavior and seasoning, chamber wall effects, first-wafer effects, air leaks, and plasma product distributions, it provides a good method to improve fault detection and classification.

In a previous study, FTIR exhaust gas monitoring was used for FDC [3]. A series of standard runs were compared to runs in which faults were purposely introduced or occurred fortuitously. Multicomponent gas mixtures containing CF₄, C2F₆, SiF₄, COF₂, CHF₃, and CO were simultaneously identified and quantified. A series of 50 etch experiments was performed on Lam's RDTB, described earlier. The faults included: incorrect chucking of the wafers (back pressure); incorrect plasma powers (TCP or bias); incorrect CF₄ pressure on the mass flow controller; regulated air leaks; and wet wafers.

Effluent management. The exhaust gases of semiconductor processes contain toxic and greenhouse gases not suitable for release into the atmosphere. Abatement systems, which are employed to reduce or collect the problem species, are expensive to operate. The FTIR multigas analyzer used in the exhaust duct before and after the abatement system provides a good way to determine what species need to be reduced and to select, qualify, control, and validate the effectiveness of the abatement system.

A thermal oxidizer treating the effluent from a rotary concentrator was tested to determine the effectiveness of a VOC abatement system. The primary effluents treated were propylene glycol methyl ether acetate (PGMEA), isopropyl alcohol (IPA), and methyl alcohol (MeOH). The concentrations of each, relative to the maximum measured concentration at the inlet and outlet of the thermal oxidizer, were analyzed. The relative concentrations of CO- and NO-regulated combustion products were also determined. An FTIR analyzer can be used to measure the destruction efficiency of the abatement system, as well as the potential production of secondary pollutants.

Cost/benefits

Each application's efficacy must be judged on its particular cost/benefit analysis. While savings due to reduced tune-up time, real-time fault detection, and others are difficult to quantify, several benefits of absorption-based process control can be easily calculated. The effect of implementing active end-point control for timed chamber cleans has the direct benefit of reducing consumables.

COO calculations should take all of the cost factors into account. These have been performed for selected APC applications [7-9]. (Information on COO calculations can be found on the Integrated Measurement Association Inc. web page [10].)

Conclusion

We have presented the potential impact of exhaust gas analysis of semiconductor processes. Based on the data, we conclude:

1. FTIR-based exhaust gas monitoring provides a reproducible, rapid, and quantitative measurement of a variety of compounds produced during etch, deposition, and chamber clean.
   
2. The mix of compounds and their time-evolution can be used for a variety of applications from R&D to production.
   
3. In many production applications, only one or at most a few gases need be monitored. Thus, low-cost NDIR sensors can be used for both process control and fault detection.
   
4. IR absorption spectroscopy offers advantages over OES spectroscopy in its ability to provide quantitative gas compositions with less drift due to window deterioration, source drift, and detector drift. IR absorption spectroscopy is applicable to thermal processes and remote plasma applications where OES cannot be employed due to lack of signal.
   
5. IR absorption spectroscopy offers advantages over RGA analysis in providing more quantitative gas compositions with little or no calibration required. IR is applicable to a wider range of pressures and is more resistive to corrosive environments.
   
6. The cost savings associated with implementing these technologies can be large.
   

Acknowledgment

This work was supported in part under National Science Foundation grant DMI-9761057 and Department of Defense contract DSWA01-97-M-0330. The authors thank Ricky Marsh of Lam Research Corp. for help with the installation and facilities support at Lam. INDUCT and Process Sense are trademarks of On-Line Technologies Inc. For more information, see the company's web site at www.online-ftir.com.

References

1. L. Zazzera, W. Reagen, in A Partnership for PFC Emissions Reductions, pp. 55-71, presented at Semicon Southwest 96, 1996.
   
2. L. Zazzera, W. Reagen, P. Mahal, in A Partnership for PFC Emissions Reductions, pp. 81-85, presented at Semicon Southwest 96, 1996.
   
3. P.R. Solomon et al., Proceedings of the Advanced Process & Equipment Control Program Workshop IX, Lake Tahoe, NE, Paper 47, Sept. 20-24, 1997.
   
4. P.R. Solomon et al., SPIE Proceedings, Vol. 2366, pp. 156-165, 1994.
   
5. M. Serio, "Adaptive Control of Chemical Processes Based on Fourier Transform Infrared (FT-IR) Spectroscopy and Artificial Neural Networks," Final Report to National Science Foundation Small Business Innovation Research Program, Nov. 11, 1998.
   
6. P.R. Solomon et al., Future Fab International, Vol. 6, pp. 245-252, 1999.
   
7. For information about the OEE calculator, contact V.A. Ames, Sematech, 2706 Montopolis Drive, Austin, TX 78741-6499, ph 512/356-7611.
   
8. P. Rosenthal, D. Dance, W. Aarts, "Estimating the Costs, Benefits, and Return on Investment of Integrated Semiconductor Process Metrology," presented at the 1998 meeting of the Integrated Measurement Association, Vail, CO, 1998.
   
9. D. Dance et al., Proceedings of the SEMICPAC Conference, Jan. 24-Jan. 27 1999, San Antonio, TX, to be published.
   
10. www.Integratedmeasurement.com.

    For further information, contact Peter A. Rosenthal, On-Line Technologies Inc., 87 Church St., East Hartford, CT 06108; ph 860/291-0719, e-mail [email protected].