Accurate metrology of epi pattern shift
06/01/2000
overview
Epitaxial deposition of silicon is widely applied in the fabrication of linear and mixed signal ICs. However, the ability to align to wafers after epitaxial-layer growth and overlay performance has always been a challenge. Furthermore, epitaxial thickness uniformity and epitaxial shift-induced distortions are important parameters that must be understood and controlled. Until now, there were no easy and reliable metrology tools or methodologies to help process engineers evaluate and optimize an epi process. Now, a unique application of a lithography stepper has yielded a very suitable method.
Peter Cheang, Keith Best, ASML Special Applications, San Jose, California
 Figure 1. Typical epitaxial growth process that results in "pattern shift" (last step). |
Measurement and characterization of the uniformity and depth of thick epitaxial layers has traditionally been based more on experience and informed guesswork than a proven, automated technique. Although the industry has moved away from relying on an experienced operator's eye to provide a reasonable estimation, the standard measurement technique of studying optical vernier structures only provides an accuracy of 0.05mm at best. Until recently, no fully automated and repeatable metrology techniques have been developed to characterize epitaxial films with thicknesses >15mm, such as those used on linear ICs using bipolar and BiCMOS processes, where epitaxial film thicknesses can reach up to 60mm.
Investigating the overlay performance of a wafer stepper for thick epitaxial layers, we have determined that the basic performance of this family of tools provides a systematic and repeatable approach for accurately determining the characteristics and uniformity of such films.
Understanding the process
For overlay and alignment, thick epitaxial layers present a number of challenges to the process engineer. Dark field alignment systems usually cannot operate through thick epitaxial films and, while pattern recognition-based systems may be able to align through thick films, their accuracy and repeatability is questionable. Even optical metrology tools have difficulty with this type of wafer, as the standard box-in-box structure is often badly distorted and the contrast is very low, especially under high magnification.
 |
Growth rates are a key factor in characterizing thick epitaxial layers. In addition to the effects from the epitaxial reactor itself and actual process conditions, growth rates differ significantly on different crystal planes. While <100> crystal orientation essentially delivers linear growth, <111> wafers do not. Consequently, features that have been defined on the silicon surface can have different growth characteristics. In fact, as previous investigations studying pattern shift and pattern distortion have shown, the geometry of a feature can also influence growth rate, with individual feature edges growing at varying rates [1, 2].
The result is a lateral displacement known as "epitaxial shift" or "pattern shift" that can significantly affect the relative position of features (Fig. 1). This shift, which has been reported to be between 0.4 and 0.9mm for every micron of epitaxial film grown, also depends on the thickness of the epitaxial layer [2]. Severe epitaxial shift can result in alignment markers being completely washed out. As the thickness of epitaxial films rises to deliver enhanced power performance, this effect will impose tremendous restraints on IC design rules, as overlaying subsequent layers becomes increasingly difficult.
Previous investigations to characterize pattern shift have been based on manual measurements, so the results provide only limited information on the uniformity and repeatability of the epitaxial growth process. Clearly, wafer processing needed a methodology that allows automatic thickness uniformity measurement and a test procedure that provides full analysis of epitaxial film characteristics. Such a method would help deliver consistent and accurate overlay performance through thick epitaxial layers and offer process engineers data to help understand process capabilities fully and allow process improvement. Ultimately, this could translate into tighter design rules, better device performance for ICs with thick epitaxial films, and a significant improvement in yield for bipolar and BiCMOS processes.
Stepper technology for characterization
Reacting to a request from a semiconductor manufacturer, we investigated overlay performance for thick epitaxial layers using the PAS 5500/100D lithographic stepper [3]. Our test used 20 silicon wafers, half with a crystal orientation of <100> and the other half <111>. Zero and first-layer phase grating alignment markers were first exposed by the stepper and were etched to 1200Å deep. Then, epitaxial layers ranging from 15-35mm thick were grown on both crystal orientations using a commercially available vertical epitaxial reactor.
 Figure 4. Theoretical alignment marker positions without process-induced distortions. |
We then used the stepper to align through the thick epitaxial film and photoresist. A total of 16 fields with 121 data points/field were measured from each of the wafers. Each of these data points were measured 25 times to evaluate the repeatability of the procedure and the stability of the measured data. The data were then analyzed with in-house software specifically developed for this test. The test data confirmed that for a <111> crystal orientation, alignment marker position variation (i.e., "epitaxial shift variation") clearly depends on the thickness of epitaxial layer grown, while for <100> wafers the effect is negligible. This is clearly highlighted in Fig. 2, which shows the alignment markers before and after the growth of a 35mm epitaxial layer for a <111> wafer.
 Figure 5. Alignment marker positions as measured on a 35mm-thick epitaxial film, using an ASML PAS 5500/100D stepper. |
Figure 3 highlights the dependency of x and y overlay on epitaxial film thicknesses. Although an overlay figure of over 1mm may in any other process seem catastrophic, these initial values closely matched expectations.
To highlight the effect of epitaxial shift, we generated a graphical representation of theoretical alignment marker positions without any process-induced distortion (Fig. 4). Then we plotted similar graphical representations using the test data as measured by the stepper. Each data point represents the measured position of one alignment marker, where each marker is 2.16mm apart. Figure 5 shows an example of the graphical representation of the distortion effect on the wafer. Additionally, similar patterns —"fingerprints" of this particular epitaxial reactor — can be seen on all wafers.
Modeling overlay with epitaxial shift
To help understand the epitaxial process and develop the test software, we also calculated the measured overlay using a rearranged version of the following empirical formula:
ETU = MO/(T x ESF)
where ETU is epitaxial thickness uniformity, MO is measured overlay, T is epitaxial film thickness, and ESF is epitaxial shift factor.
By determining the MO value for this specific case using supplied figures of 4% thickness uniformity and a 0.8mm shift/micron epitaxial layer grown, we predicted an overlay of 1.12mm for a 35mm epitaxial film thickness. Comparing this value to data in Fig. 3, the theoretical overlay value and measured overlay value are very close for a 35mm film thickness.
Marker quality and repeatability
 |
To check the repeatability of this technique, we measured each marker 25 times. On all wafers tested, the average mark quality was higher than that required by our stepper for overlay measurement. Overall, the test achieved repeatability on all wafers of better than 4nm (1s). In fact, with the multiple correlation coefficient (MCC [4]) values on all wafers in the high 90% range, the correlation between the modeled overlay data and measured data was very high (see table).
Reactor substrate orientation, wafer position
While the prime goal in this investigation was determining the overlay performance of our stepper for thick epitaxial layers, our data also allow growth rates and epitaxial layer uniformity to be characterized according to wafer position in the epitaxial reactor. There was a clear difference in a number of wafer-correctable values between wafers at the top of the reactor and those in the bottom position. In fact, position in the reactor also influenced the average overlay value in this particular investigation; wafers on the bottom of the reactor showed significantly better standard deviation distributions than those at the top.
Conclusion
Epitaxial film characteristics depend on a wide range of process variables such as the substrate orientation, epitaxial reactor configuration and processing conditions. We have shown how a stepper can be used as an automatic and accurate overlay and metrology tool for thick epitaxial layers. Along with ensuring overlay for subsequent process steps, this also provides process engineers with valuable data to characterize and optimize the epitaxial process further.
Given the very promising results from this initial investigation, work will continue in characterizing thick epitaxial growth using lithography tools. We are already investigating accurate determination of epitaxial shift factor and thickness uniformity of the epitaxial layer. This will not only lead to a greater understanding and help optimize the thick film epitaxial process, it will also benefit thinner (2 to 4mm) epitaxial layers.
References
- C.M. Drum, C.A. Clark, JCES, Vol. 117, p. 1401, 1970.
- S.P. Weeks, Solid State Technology, p. 111, Nov. 1981.
- ASML PAS 5500/100D is a commercially available i-line stepper for high-throughput sub-half micron production with a 0.48-0.6 variable numerical aperture lens. While ASML is focused on developing lithographic solutions, ASML Special Applications' primary concern is with the processes involved in IC fabrication. This group studies applications that can benefit from the performance of ASML's tools, including undertaking investigations such as using the PAS 5500/100D to characterize thick epitaxial layers.
- 4.EThe ASML alignment marker consists of two phase-grating structures with two different pitches, 8.0mm and 8.8mm in both x and y directions. The MCC values, ranging between 0 and 1, show how well the modeled data are related to the measured data.
Peter Cheang received his BSEE from National Taiwan University, his MSEE from UC Santa Barbara, and his MBA from Santa Clara University. He has more than 12 years of semiconductor experience in various process-engineering and marketing roles. Cheang is product-marketing manager at ASML Special Applications, 1971 N. Capitol Ave, San Jose, CA 95132; ph 408/719-6382, fax 408/719-6378, e-mail [email protected].
Keith Best received his BS with honors in materials science from the University of South East London. He has more than 14 years of semiconductor experience in various process engineering and application roles. Best is product support manager with ASML Special Applications.