Using broadband reflectometry for fast trench-depth measurement
02/01/2003
by Robert Herrick, Teina Pardue, Fairchild Semiconductor, West Jordan, Utah Phillip Walsh, n&k Technology, Inc., Santa Clara, California
overview
Accurate, fast monitoring of trench depths at early stages of production is essential for minimizing production cost and eliminating waste. Conventional methods are available, but increasingly the need is for real-time metrology. In one fab, engineers have applied a specific model of broadband reflectometry to quickly check trench depths on product wafers.
Trench depth is a critical parameter affecting performance characteristics in many current semiconductor applications, including shallow trench isolation and power discrete vertical MOSFETs. Current methods for measuring trench depth, such as atomic force microscopy (AFM), profilometry, and cross-section scanning electron microscopy (SEM), are not always well suited to this task. SEM is destructive, and both SEM and AFM are time-consuming, making these methods unsuitable for real-time monitoring of trench properties in the production cycle. Profilometry is not as accurate as other methods and typically requires test structures that are wider than product trenches so the stylus fits into the trench.
We have successfully applied broadband reflectometry, with the n&k 3000 TMS metrology tool, to real-time nondestructive measurement of trench parameters. This form of broadband reflectometer is based on toroidal mirrors, without lenses or beam splitters [1]. The result is spectrophotometric information over a wide range of wavelengths — from deep UV to near IR — and a high signal-to-noise ratio over the entire wavelength range. This makes the instrument extremely sensitive to small changes in sample conditions. The reflectance data is analyzed using the Forouhi-Bloomer dispersion equations (the "n&k method" [2]) to determine optical properties and film thickness for any number of film layers. This instrument yields uniformity maps of full, patterned wafers and has been used in wafer fabs for monitoring film thickness and optical properties [3, 4].
Application to trench measurement
Geometric structures in a device wafer cause distinct features in the measured spectrum, distinguishing the sample from one having uniform film structure. Variations in the geometry, such as changes in trench depth and width, can be tracked through changes in these features.
Although the n&k tool may generate a trench profile through rigorous wave analysis of the measured spectra, our emphasis is on enabling the monitoring of drift in an etch process. The effect of a critical trench parameter, such as depth, on the measured spectra is isolated from changes in other parameters such as mesa film thickness, after which variations in the parameter can be tracked with extraordinary resolution. Although a single, isolated measurement might be subject to some interpretation, variations of the measured parameter are completely unambiguous. The only task remaining is a simple calibration step that is easily achieved with a small set of pre-staged samples.
 Figure 1. Measurement results for a) mesa oxide thickness and b) silicon-to-silicon trench depth (all dimensions in Å). |
null
To verify the capabilities of the tool, we spent some time focusing on systematic variations in etch time, which will typically show a strong correlation with trench depth. The n&k method has performed as well as — perhaps better than — other methods, such as SEM, AFM, and profilometry, in its ability to track changes in trench depth. There has also been success in tracking changes in other process conditions, such as focus and energy dose of the light source during photoresist exposure, but we do not discuss these results in this article.
The metrology tool's trench measurement returns the average trench depth over the entire measurement spot size of ≤50µm for the present configuration. Since the spot size is small compared to typical die width but large compared to a typical trench structure, the technique can be used to map trench depth patterns across a single die, without being adversely affected by fluctuations in trench depth between individual trenches.
One other consequence of this approach is that the thickness of films on trench mesas, and even inside trenches, falls right out of the analysis. Trench parameters are thus obtained simultaneously with the thicknesses of films inside and outside trenches.
 Figure 2. Comparison of AFM, profilometer, and n&k 3000 TMS measurements (six wafers with etch times increasing in 5-sec intervals, beginning at 90 sec). |
null
A simple application
Various steps in device fabrication may involve etching through films on a wafer to define where trenches will be etched. The resulting structures have thin films on the mesas. As mentioned above, any film structure already characterized by n&k broadband reflectometry can be readily incorporated into a trench measurement with both measurements done simultaneously. As an example, we looked at a 200mm oxide covered silicon wafer etched with ~1.2µm-deep trench structures. The trench depth and mesa oxide thickness distributions across the wafer were determined simultaneously in only a few minutes.
In this case, we measured 69 dies, each die being measured at its center on the actual product area. The results are shown for mesa oxide thickness and trench depth in Fig. 1. Our data for trench depth are silicon-to-silicon depth, without the oxide contribution. The entire scan of 69 points was completed in ~3 min.
The measurement recipe is fast and easy to set up. A typical calibration sequence involves the optical characterization of each type of film involved in the structure on a separate, polished, uniformly coated wafer, along with a short series of real samples that vary only in etch time, and span a typical process window. The trends in trench depth and wafer uniformity are obtained without any further calibration, although typically one or more of the samples is measured using some other technique for calibration.
The recipe for a new product is usually implemented and calibrated in a few minutes. More important, the recipe can be easily generated in-house by process engineers. Thereafter, a new product is monitored by simply loading a wafer and selecting the recipe.
 Figure 3. SEM (dashed lines) vs. n&k data (dotted lines) from trench depth measurements on four wafers. |
null
Comparison with AFM
We etched six wafers to fabricate deep 4.5–5.5µm linear trenches, varying trench depth slightly from wafer to wafer by varying etch time. Each wafer was measured at five points using AFM, a stylus profilometer, and the n&k 3000 TMS (Fig. 2). The additional set of data in the figure accounts for the constant offset between the AFM and n&k measurements. This offset calibration is typically the only modification required to accommodate a new nominal trench structure into a recipe. Similar results have been obtained with trench depths ranging from 0.4–9.0µm.
Data in Fig. 2 show that trench depth measurements using the n&k broadband reflectometry method are comparable to AFM, which is arguably the most accurate mechanical measure of trench depth. Compared to AFM, the n&k reflectometry method requires only ~3 sec/measurement.
null
Comparison with SEM
We etched four 150mm silicon wafers with linear trenches of varying depth and then cross-sectioned them so we could measure the center die of each using SEM. We scanned the other half of each wafer horizontally, using the broadband reflectometry tool's rapid measurement capability. The central die in these scans closely correspond to the results of SEM measurements due to their proximity to the SEM measured sites. The sheer number of die measured demonstrates the high throughput of the method, as well as its ability to obtain uniformity information across a product wafer. Each scan consisted of nearly 60 die measured in ~3 min.
Measurements are shown in Fig. 3, where the dashed horizontal lines represent the SEM center die measurements. The corresponding broadband reflectometry measurements show good correlation with SEM at the center points, while the extended distributions demonstrate the ability to reveal center-to-edge trench depth variation across wafers at very high resolution (Fig. 4).
null
Repeatability, reproducibility
We conducted gauge studies on both AFM and broadband reflectometry measurements. Ten wafers with a nominal trench depth of 1.65µm were measured three times by each of three appraisers on each tool. Measurement variations were calculated using the average and range method. Figure 5 lists the repeatability (equipment variation), reproducibility (appraiser variation), and the combined repeatability and reproducibility (R&R) values as a percent of the total variation of the study (i.e., the lower the value, the better the performance of the tool).
One key advantage that the broadband reflectrometry method has in gauge studies is that there is no "art" involved in performing a measurement. The only operator variation is in sample placement, an effect that is largely eliminated through the use of pattern recognition and robotic handling capabilities.
Production applicable
Overall, we found that the n&k-method of broadband reflectometry is a viable alternative to slower, destructive methods for real-time monitoring of trench depth in a production environment. This method demonstrated resolution as good as, or even better than, AFM. Its speed of measurement and nondestructive nature mean that in practice every wafer can be measured, drift in production parameters is caught more quickly, and the effects of adjustments in the production cycle can be observed in real time.
The fact that measurements can be done on a real transistor means that there is no need to accommodate additional test structures on a product wafer for measuring trench depth. The additional overhead involved in recipe generation is minimized, typically requiring only a simple calibration step to accommodate new structures. Alternately, the calibration step can be skipped completely if only variations in trench depth are of interest.
It also helps that the method is capable of standard thin-film analysis in addition to trench characterization, so it could potentially replace multiple in-line measurement tools.
References
- D. Buerman, et al., US Patent No. 6,128,085., Oct. 1999.
- A.R. Forouhi, et al., Phys. Rev. B, 38, 1865, 1988.
- G.G. Li, et al., Semiconductor Fabtech, pp. 279–283, Oct. 1998.
- C. Bencher, et al., Solid State Technology, pp. 109–114, March 1997.
Robert Herrick received his BS in physics and MBA from Brigham Young University, Provo, UT. He is an R&D process engineer at Fairchild Semiconductor, 3333 W 9000 S, West Jordan, UT 84088; ph 801/562-7541, [email protected].
Teina Pardue is working toward her BS at University of Phoenix. She is a process-engineering technician at Fairchild Semiconductor.
Phillip Walsh received his PhD in physics at Louisiana State University. He is a senior applications engineer at n&k Technology.