Measuring deep-trench structures with model-based IR
03/01/2006
Advanced DRAMs using deep-trench capacitors have small cell designs and excellent scaling potential. At the 90nm node and beyond, measuring the dimensions of deep trenches during fabrication is increasingly important for process control. Model-based infrared reflectometry (MBIR) provides a unique capability to measure subtle details of the trench shape that can affect device yield, such as top and bottom CD, neck depths, and recess depths. MBIR is a high-speed metrology technology with the sensitivity and spatial resolution required to control process variability in a production environment.
Michael Gostein, Peter A. Rosenthal, Alex Maznev, Philips AMS, Natick, MA Alexander Kasic, Peter Weidner, Infineon Technologies, Dresden, Germany Pierre-Yves Guittet, Inotera Memories, Taiwan
DRAMs using deep-trench capacitors make up a significant portion of the commodity and embedded DRAM markets. In deep-trench technology, the charge-storage capacitor is formed by etching a deep, high-aspect ratio cavity in a doped-silicon region, lining it with dielectric, and filling it with doped poly-silicon to make an opposing electrode. One challenge in scaling these devices, just as for competing DRAM stacked-capacitor technology, is to maintain sufficient capacitance while reducing the planar area on the chip.
For the transition to 90nm deep-trench technology, the capacitance challenge was addressed by adopting bottle-shaped trenches on a rotated substrate [1]. The anisotropic wet bottle etch increases the trench CD below the transistor region and also produces trenches with square cross section, resulting in increased capacitor surface area and higher packing density.
However, these advanced designs also introduced metrology challenges. As deep-trench devices have been scaled and packed more tightly, the bottom CD has become a more critical parameter. With the trench-profile complexity increasing with scaling, depth-dependent features such as bottleneck depth must now be monitored to ensure process control. Scaling also makes it more difficult to characterize recess structure geometry following etch-back of both polysilicon and photoresist-fill during the fabrication of the upper elements of the DRAM devices. Atomic force microscopy (AFM) profiling, the historical approach, has become increasingly challenged at advanced nodes by narrower CDs and increased aspect ratios.
Model-based infrared reflectometry
These metrology challenges have been met by Infineon and its technology partners by using model-based infrared reflectometry (MBIR), which provides noncontact, rapid measurements with the sensitivity to extract details of complex trench profiles [2, 3]. MBIR uses an infrared beam with wavelengths >1.4µm to probe the trench structure. At these wavelengths, silicon microstructures are transparent. The MBIR probing beam thus penetrates the entire structure and generates an interference pattern associated with reflections from interface layers. The interference pattern exhibited by the infrared reflectance spectrum encodes details of the trench shape that are extracted using model-based analysis algorithms. Here we highlight applications of MBIR to measure top and bottom CD, neck depth, and recess depths in Infineon’s deep-trench structures. Exact structure dimensions and wavelength scales have been omitted from all relevant figures and text to protect proprietary process information.
Measurements were performed using MBIR tools from Philips AMS (model IR3000). These tools have high photometric accuracy, achieved by employing a proprietary optical design to suppress stray wafer back-side reflections, a high-sensitivity linearized detection method, high accuracy positioning, autofocus, and automatic drift correction. The resulting accuracy and repeatability allow collection of spectra that encode profile information that can be captured through model fitting. The IR3000 has a spot size of 200×600µm, typical measurement and analysis time of 2-3 sec., and wavelength range of 1.4-20µm. The tool is built on a GEM-compliant automated platform designed for high volume in-line process control applications.
In contrast to visible light measurements, for MBIR the measurement wavelengths are significantly longer than the DRAM array pitch (typically <0.35µm). As a result, light scattering is minimized and DRAM trench structures can be modeled as multilayered film stacks with optical properties (n and k) of the various layers computed according to effective medium approximations (EMA). These EMA models determine the local optical constants by appropriately combining the optical properties of the silicon, air, and various fill materials used in the structures [4]. With this approach, the trench profiles are represented as simple layered optical models, as shown in Fig. 1. The left side of each sub-figure shows a cross-sectional schematic of a portion of an array corresponding to different deep-trench structures; the right side shows the equivalent optical model, in which trenched layers are represented as voided or filled materials modeled using an effective medium approximation. The gradual change in CD from top to bottom in Fig. 1a is represented by graded n and k with differing top and bottom void fractions, while the abrupt change in CD in Fig. 1b at the bottom of the bottleneck is represented by using two layers of different void fraction with a sharp interface.
Top and bottom CD measurements
CD measurement following the initial deep-trench etch is especially important because CD strongly correlates with yield and defectivity. While CD at the top of the silicon nitride mask layer can be measured nondestructively using CD-SEM and other means, CDs at the nitride/silicon interface and near the trench bottom are more difficult to measure because the regions of interest are buried in the silicon and inaccessible to other nondestructive techniques. MBIR provides an effective means to measure these parameters nondestructively and at high throughput, greatly reducing the need for sampling with cross-sectional SEM.
Figure 2a shows an MBIR spectrum of a typical deep-trench structure with depth between 5-8µm and CD in the 50-200nm range. The effects of various structure dimensions on the spectrum shape can be described in simplified form. The oscillation pattern of the spectrum arises from the interference of reflections from the top and bottom of the trench layer, and the fringe period scales inversely with the total trench depth. The top CD has a strong effect on the overall reflectivity, since increasing top void fraction lowers the optical contrast between air and the top of the sample, lowering reflectivity. Finally, the bottom CD affects the modulation depth of the interference fringes, since an increase in bottom void fraction increases the optical contrast between the trench layer and the silicon below. The exact spectrum, taking into account these parameters and other structure details, can be calculated using Fresnel reflectivity equations using the optical model depicted in Fig. 1a.
Best-fit values for the top and bottom void fraction parameters are determined by iteratively fitting the measured spectrum. The model fit (red line in Fig. 2a) agrees well with the measured data. The top and bottom CD values are computed from the fitted void fraction parameters using simple calibrations. Dynamic repeatability of top and bottom CD measurements is typically <0.5% (1σ) and <0.25% (1σ), respectively.
High repeatability and short measurement time allow MBIR to be employed to generate full-wafer uniformity maps. These give new insight to the process engineer, who previously had to rely on destructive SEM sampling of a limited number of points. For example, Fig. 3a shows a 156-dice map of bottom CD on a 300mm wafer, with CD values normalized to the mean. The map shows some areas with significant deviations. These types of deviations can be correlated with defectivity, and they provide valuable information for yield learning, process troubleshooting, and optimization.
Figure 3b shows the correlation of the MBIR results with SEM measurements on selected dice for three wafers. SEM measurements are the average of 3-4 individual trenches on each die and have a statistical error of ±9% (1σ) due to trench-to-trench variability, while MBIR results average over thousands of trenches within the measurement spot. Despite the differences in sample size and limits to the site correspondence, Fig. 3a shows good overall correlation between MBIR and SEM, and validates the interpretation of the uniformity map.
Bottle-shaped trenches
Bottle-shaped trenches used at 90nm and below are created by subjecting the bottom portion of the trench to isotropic etching while the top portion of the trench is covered with a passivation layer. The bottleneck is typically about 1μm deep, with CD 20-40% larger below the neck than above it, as illustrated in Fig. 1b. Just as for bottom CD, it is difficult to measure the neck depth by conventional nondestructive means.
MBIR provides an effective solution. Figure 2b shows the MBIR spectrum from a typical bottle-trench structure. In contrast to Fig. 2a, there are two oscillation patterns. The one with smaller period arises from the interference of reflections from the tops and bottoms of the deep trenches, while the one with larger period arises from the interference of reflections from the tops and bottoms of the bottlenecks. Simulation of the exact spectrum using the model depicted in Fig. 1b allows best-fit determination of the neck depth. The model fit (red line in Fig. 2b) shows good agreement with the measured data. Neck-depth dynamic repeatability is typically <0.1% (1σ) [3], and results correlate very well with cross-sectional SEM data, as shown in Fig. 4a.
 Figure 4. MBIR correlation vs. SEM data for deep trench measurements of a) bottleneck depth and b) recess depth. |
Using nondestructive MBIR measurements and high density sampling, Infineon realized a significant improvement in the control of bottle-etch uniformity. Previously, standard metrology equipment and SEM cross-sections alone had not provided enough high-quality data to master the process [2].
Recess depths
Following fabrication of the deep-trench storage capacitor, the transistor and isolation structures are formed near the top of the deep trench. This process involves several cycles of trench fill and etchback to form recess structures. In three successive stages, the recess depths are typically 500-1000nm, 50-300nm, and 20-100nm below the silicon. Figure 1c shows an example of a recess structure cross-sectional schematic.
Recess depth measurements pose considerable measurement challenges. AFM measurements are problematic for the first and deepest recess structures, where high aspect ratio prohibits probing with available tips. Subsequent shallower recess steps can be monitored with AFM technology in current nodes, but a trend to smaller CDs at constant aspect ratios poses scaling hurdles at 65nm and beyond that require further tip development.
Measuring recess structures with MBIR is similar to measuring a shallow trench [5]. Figure 2c shows a typical reflectance spectrum from a recess structure. The low-period oscillation arises due to interference of reflections from the interfaces at the tops and bottoms of the unfilled part of the recessed structures, and the oscillation period is determined by the recess depth and CD. The data are fit using the layered optical model depicted in Fig. 1c. Typical recess 1 measurement dynamic repeatabilities are <0.2% (1σ). MBIR again correlates well to SEM for a split lot of recess 1 structures (Fig. 4b).
Optical measurements with MBIR provide significant throughput and cost-of-ownership advantages versus AFM, SEM cross-sections, and FIB measurements. In addition, as device geometries scale, AFM tip development becomes more difficult, while MBIR measurements are unaffected. Furthermore, recent work shows that MBIR can measure depth below silicon, while AFM only measures overall depth. Accuracy studies are ongoing to establish the lower limit of MBIR capability on very shallow structures.
Conclusion
As deep-trench capacitor DRAM devices are scaled to smaller geometries, metrology plays a growing role in the process control strategy. Measurement of the bottom CD of the deep trench becomes more critical, because this parameter is strongly associated with yield and defectivity. The use of bottle-shaped trenches has introduced vertical structure into the trench profile that must be measured to ensure process control. Recess dimensions are also becoming harder to measure using historical methods due to tighter CDs and higher aspect ratios.
Model-based infrared reflectometry offers a new capability for cost-effective development and production of advanced DRAM structures. It offers an approach to measuring a wide range of structure parameters, and can offset or replace the use of more time-consuming and expensive techniques. Furthermore, MBIR’s high repeatability and high throughput enable the routine measurement of within-wafer uniformity and wafer-to-wafer process variations that previously were only measured with isolated sampling. It thus provides more complete data for process optimization and control while lowering overall production costs.
Acknowledgments
The authors would like to thank Alex Mazurenko, Carlos A. Durán, Han C. Chang, Jonny Hoglund, Greg Merklin, Paramjeet Gulshan, Tony Bonanno, and Chris Moore of Philips AMS for assisting with measurements and data analysis, and for technical discussions.
References
- P.S. Parkinson, et al., “Novel Techniques for Scaling Deep Trench DRAM Capacitor Technology to 0.11µm and Beyond,” 2003 International Symposium on VLSI Technology, Systems, and Applications, pp. 21-24.
- U. Mantz, A. Kasic, “Model-based Infrared Spectroscopy: New Opportunities for In-Line Process Control,” Future Fab, Vol. 19, June 2005.
- P. Rosenthal, et al., “Model-Based Infrared Metrology for Advanced Technology Nodes and 300 mm Wafer Processing,” in Characterization and Metrology for ULSI Technology 2005, ed by D.G. Seiler, AIP Conference Proceedings, pp. 620-624, 2005.
- G.E. Jellison, “Physics of Optical Metrology of Silicon-based Semiconductor Devices,” Handbook of Silicon Semiconductor Metrology, ed by A.C. Diebold, Marcel-Dekker, 2001, pp. 723-760, 2001.
- S. Zaidi, et al., “FTIR-based Nondestructive Method for Metrology of Depths in Polysilicon-filled Trenches,” Proceedings of the SPIE, Vol. 5038, pp. 185-190, 2003.
Dr. Michael Gostein is technology manager at Philips Advanced Metrology Systems, 12 Michigan Dr., Natick, MA 01760; ph 512/231-2295, e-mail [email protected].
Dr. Peter A. Rosenthal was formerly technology manager at Philips AMS, and is now director of engineering for precision positioning at Zygo Corp.
Dr. Alex Maznev is senior scientist at Philips AMS.
Dr. Alexander Kasic is system expert with Infineon Technologies, Dresden, Germany.
Dr. Peter Weidner is senior staff engineer at Infineon Technologies.
Pierre-Yves Guittet is metrology section manager at Inotera Memories, Taiwan.