Advanced process control for epitaxial silicon
09/01/1998
COVER ARTICLE
Advanced process control for epitaxial silicon
Weimin Zhang, Matt Richter, Peter Solomon, ON-LINE Technologies, East Hartford, Connecticut
Yiorgos Kostoulas, Gerhart Kneissl, ADE Corp., Westwood, Massachusetts
Wim Aarts, Wacker Siltronic Corp., Portland, Oregon
Ann Waldhauer, Applied Materials Corp., Santa Clara, California
A precise and accurate Fourier transform infrared spectroscopy-based epitaxial silicon characterization technique provides information on the doping profile as well as epi layer thickness. This comprehensive, model-based approach results in consistent, machine-independent measurements that will significantly ease the application of Fourier transform infrared spectroscopy-based advanced process control algorithms and computer-integrated manufacturing technology.
Process control issues are becoming increasingly important as the semiconductor industry switches to larger, more expensive 300-mm wafers. Larger wafers make the use of test wafers less economical and increase the need for improved metrology. Ideally, tight process control for epitaxial silicon requires the ability to perform on each product wafer accurate [1], repeatable [2], and reproducible [3] multipoint measurements. Other requirements include a noncontact method and the capability for fast measurements in order to maintain high throughput.
Epitaxial silicon layers are currently produced using timed recipes. The process is monitored by statistical sampling (typically 1 in 25 wafers) followed by off-line measurements. Tighter process control will require some type of feedback using in situ or in-line measurement of each wafer, preferably in a multipoint format. Measuring each wafer will instantly detect problems during deposition (e.g., hydrogen interruptions) and collect data on each wafer. Measurements on each wafer can also ensure that the process is not beyond equipment operation specification limits (out-of-spec), providing improved quality assurance and control. The optimal placement of process-monitoring equipment is on the fabrication tool, as it limits extra wafer handling. In-line monitoring requires the use of equipment that is reliable, fast, robust, and easily integrated into a cluster tool. When the time allowed for an in-line measurement is limited, or more extensive information is necessary, off-line systems allow more complete analysis, such as fine mapping of the wafer epi thickness. These capabilities are especially useful to "tune-up" deposition chambers for uniformity, and will become increasingly important for the upcoming 300-mm wafer market.
In the case of epitaxial silicon, optical methods using Fourier transform infrared spectroscopy (FTIR) have been used extensively to measure film thickness [4]. These methods are noncontact and relatively fast. Their current commercial implementations have fundamental drawbacks, however, which limit their accuracy and precision, especially for thin epi [5]. The most significant barrier to the use of FTIR as a process control sensor is the lack of intermachine precision that results from today`s nonmodel-based analysis routines. A new model-based FTIR method is now available, with high precision (including intermachine precision) and accuracy (including the case of thin films) [5], which provides additional information about the doping profile.
New model-based FTIR methods
Table 1 compares the main features of the new method [5] with commercially available FTIR techniques. While all techniques share the same physical principle (i.e., how infrared light is reflected from an epi wafer), they differ on the use of the extracted information. The current commercial methods (the interferogram methods) [4] analyze only a limited number of features from the interferogram, namely the distance between the "centerburst" and the "sideburst"; the new model-based approach, on the other hand, uses the entire frequency range available to model the experimental data.
 |
This approach to epi layer thickness measurement is complemented by advanced and improved commercial FTIR spectrometers [6]. These systems allow fast measurement times, which are critical for wafer mapping. Some of these new systems, because of their compactness, can also be easily integrated into cluster tools as well as off-line systems.
We tested the new FTIR epi thickness-monitoring systems for their precision (repeatability and reproducibility), detection limit, accuracy, and in-line and off-line applicability (Table 2).
 |
Measurement precision (repeatability). The precision for epi layer thickness is defined as the standard deviation (1 s) for 25 repeated measurements at the same spot on a wafer, with the wafer unloaded and replaced in the cassette after each measurement. Consequently, the precision includes the reflectance measurement precision as well as the robot-positioning precision. The new systems yielded 1-s values of 0.4 nm for >1-?m epi films, and a substrate doping of 8 ? 1018 cm-3. The 1-s values increased to 6 nm for thinner epi films due to the limited number of fringes measured within the observed spectral range and the higher susceptibility of the measurements to factors such as robot-positioning precision. These values are largely independent of the thickness range measured, but depend on the substrate doping, with the higher precision occurring for the higher doping. A =1-nm precision is typical for a substrate-doping concentration in the 1 ? 1018 - 1 ? 1019 cm-3 range. The high level of precision demonstrates the advantages of the model-based analysis over standard commercial methods (Table 2).
Intermachine precision (reproducibility). A test on four systems using the same set of three epi wafers, with epi thickness ranging from 1-5 ?m, assessed the instrument-to-instrument correlation of the new tool. We carried out the correlation test without any bias adjustment between the systems and compared the same spot on each sample. The precision (1 s) for each spot of a given epi wafer on the different machines was 3 nm, more than two orders of magnitude higher than that of current commercial systems (Table 2). This high value of intermachine precision is an extremely important characteristic because it completely eliminates the required use of "golden wafer" thickness standards that are exchanged between epi wafer suppliers and their customers.
Very thin films (detection limit). At the low thickness range (<2 ?m), interferogram-based techniques become problematic due to merging of the sidebursts with the centerburst. Subtraction of the centerburst to reveal the sidebursts has been used, but the technique is prone to errors due to the nature of the subtracted curves (steep slopes) and the requirement of reference wafers. The model-based approach, however, is immune to these problems and offers more accurate values for very thin films. The thickness for the thinnest wafers is ~0.15-0.25 ?m (depending on the substrate doping), at a precision of 6 nm. These values represent a significant improvement over the current commercial interferogram methods that have a detection limit of >1 ?m.
Accuracy. We compared our results with those from secondary ion mass spectroscopy (SIMS) to assess the accuracy of the new model-based method. From our study [5] of a large number of samples from different manufacturers and with different doping profiles, we found the model-based analysis to be accurate to 20 nm, which is within the experimental accuracy of SIMS. There was no bias, representing a significant improvement over current experimental methods that do not correlate as well with SIMS and have biases as large as 200 nm and >200 nm for thinner (<1 ?m) epi films.
Applications
Reduction of measuremment times due to faster spectrometers and more computing power, as well as higher precision and accuracy, allows the new model-based system to provide both in situ and ex situ process monitoring and control.
In-line process monitoring. An integral part of in-line process monitoring is a sensor that can provide real-time, accurate information on processed materials without hindering or altering the process. Figure 1 shows thickness data obtained with the new FTIR system mounted on the cool-down chamber of an Applied Materials Epi Centura cluster tool, which allowed measurements without any decrease in system throughput. The figure shows thickness data from three different deposition chambers, one chamber being out of tune with the other two. The faulty chamber had a small air leak that affected the thickness of the epi layer. Because the drift was present in only one of the three growth modules, statistical sampling may not have located the problem as quickly.
 |
Figure 1. In-line thickness measurement performed on three chambers. One of the chambers has some drift.
Process control. In situ, real-time process control improves equipment effectiveness by reducing process time and downtime while maintaining product quality. The most crucial requirement is a sensor that can provide real-time information on the processed materials and structures. This device needs to be sensitive to the designated parameters of interest on a time scale shorter than the time needed to modify the sample properties. In the case of an epi film, the center point thickness (measured from epi layer surface to region with 50% of doping) can be used as the feedback variable for run to run (R2R) process control software (Fig. 2). We purposely set the initial growth rate estimate for a 4.5-?m epi film while setting the controller to 6 ?m. Additionally, we introduced several process disturbances (e.g., changes in reactor temperature) to test the control algorithm. At the end of the run, a new thickness set point was introduced, and the system responded with no out-of-spec wafers. Adoption of this technology for process control has immediate payback in reduced out-of-spec wafers and waste (via instantaneous fault detection), less thickness variation within each lot as well as from lot-to-lot, and the added confidence provided by performing quality control on each and every wafer.
 |
Figure 2. Initial test results of R2R control software demonstrating fault recovery and instantaneous recipe tuning.
Off-line mapping. Off-line mapping of epi layer thickness can help speed process development and chamber matching, and is a required quality control step for current production facilities. In general, the quality assurance sampling requirement is set by the particular user, and varies from user to user, requiring a programmable instrument. A cost-effective wafer-mapping instrument must have high throughput and short measurement time. Issues such as analysis time, data acquisition time, and robotics speed become critical. The analysis time has decreased drastically with the advance of fast computers, and is no longer the limiting factor. A one-point measurement time of <1 sec can be obtained with current state-of-the-art robotics and FTIR spectrometers, giving a system throughput of 40 wafers/hr using a 25-point map.
For process development, higher-density maps can be used to investigate and identify epi layer thickness patterns, such as "bowl" shapes, "donut" shapes, etc. Thickness maps (317 points), obtained in less than 7 min for 3.8-?m (Fig. 3) and 0.8-?m (Fig. 4) epi films on 200-mm wafers, show that the thicknesses are not homogeneous across the wafers. For the 3.8-?m epi wafer, a "donut" shape is observable, with variations in thickness up to 80 nm. This type of pattern is typical in single-wafer production systems. For the 0.8-?m epi wafer, the thickness is symmetrical on either side of one diameter of the wafer, and variations in thickness of up to 90 nm can be observed. This "potato chip" shape is typical of oven-grown epi layers. For these wafers, the thickness variations represent 2-11% of the average thickness.
 |
Figure 3. Thickness map (317 points) for a ~3.8-?m epi wafer displaying a "donut" shape characteristic of a single-wafer reactor.
 |
Figure 4. Thickness map (317 points) for a ~0.8-?m epi wafer showing a symmetry along one of the wafer`s diameters. This pattern can appear on epi films grown in barrel reactors.
Conclusion
A new model-based, FTIR-based, epitaxial silicon characterization technique has been developed that is accurate, precise, and fast enough to meet most significant trends in today`s epi silicon market. This method provides a reliable and robust sensor for epi production, and also for the emerging requirements of advanced process control, combined with the ability to measure submicron films quickly and accurately. Other added benefits of this new technique include the elimination of "golden wafer" calibration standards, easing process development, transfer, and tuning, while providing more accurate metrology information for statistical process control and quality assurance. High-density thickness maps can be generated in a fraction of the time previously required, easing reactor tuning as well. The new model-based analysis also provides information on doping profiles and transition widths that had previously been obtainable only with destructive testing (SIMS and spreading resistance profiling). In-line installations of the system are capable of run-to-run process control, and real-time chamber tuning, and provide instantaneous fault detection as well as metrology data on every wafer. Based on all the potential benefits, FTIR-based epi process control should play a critical role in current and future production environments.
References
1. Accuracy is defined as the standard deviation of the correlation with secondary ion mass spectrometry (SIMS) measurements.
2. Repeatability (also called precision) is defined as the standard deviation of the measurements when the sample is loaded, measured, and unloaded 25 times.
3. Reproducibility (also called intermachine precision) is defined as the standard deviation of the measurements of the same wafer (same spot) on different machines, and defines intermachine and interlaboratory correlation.
4. Biorad, Notes No. 26, "Measurement of Silicon Epitaxial Layer Thickness Using IR Interferometry," Nicolet, FTIR Spectrophotometer.
5. S. Charpenay et al., "Model-based Analysis for Precise and Accurate Epitaxial Silicon Measurements,"Solid State Technology, Vol. 41, No. 7, p. 161, July 1998.
6. For an example, see ON-LINE Technologies` 2100 Process FTIR.
For more information, contact Yiorgos Kostoulas, ADE Corp., 80 Wilson Way, Westwood, MA 02090; ph 781/467-3767, fax 781/461-1575.