Metrology for Characterization of Wafer Thickness Uniformity During 3D-IC Processing.ndd.indd

There is a constant desire to increase substrate size in order to improve cost effectiveness of semiconductor processes. As the wafer diameter has increased from 2” to 12”, the thickness has remained largely the same, resulting in a wafer form factor with inherently low stiffness. Gravity induced deformation becomes important when using traditional metrology tools and mounting strategies to characterize a wafer with such low stiffness. While there are strategies used to try to reduce the effects of deformation, gravitational sag provides a large source of error in measurements. Furthermore, glass is becoming an important material for substrates in semiconductor applications and metrology tools developed for use for characterizing silicon are inherently less suitable for glass. Using a novel mounting strategy and a measurement technique based on optical interference provides an opportunity to improve on the methodologies utilized to characterize wafer fl atness (warp, bow) and total thickness variation (TTV). Not only can the accuracy of the measurement be improved, using an interference based technique allows for full wafer characterization with spatial resolution better than 1 mm, providing substantially more complete wafer characterization. Introduction Historically, use of glass wafers in the semiconductor industry has been primarily for MEMs and CMOS image sensor applications. These applications typically had loose specifi cations for TTV and warp. Using glass as a carrier wafer for precision thinning of silicon in 3D-IC applications requires that the thickness uniformity and warp are tightly controlled since non-uniformities in the carrier directly impact the accuracy of the silicon TTV. Another challenge is given by the fact, that over the past several years, wafer diameters have increased dramatically; resulting in the requirement to accurately characterizes extremely high aspect ratio wafers (300 mm diameter and thickness < 1 mm). High aspect ratio parts have inherently low stiffness and characterizing the fl atness of such a component is extremely challenging due to gravity effects. Conventional mounting methods, e.g. three/four-point mounts, are less suitable for fl atness characterization of such high aspect ratio parts due to a great deal of defl ection of gravity and sensitivity to how the wafer is placed on the mount. This leads to erroneous results when trying to characterize warp and bow. Corning® Tropel® has developed a novel distance measuring interferometer based on a frequency stepping laser that is well-suited to characterize the fl atness and TTV of glass wafers. In fact, several commercial interferometers capable of characterizing the fl atness, thickness, and TTV of 300 mm diameter glass wafers have been installed. In addition to novel mounting strategies that substantially avoid errors given by historical techniques; this metrology tool has extremely high accuracy and a tight pixel density. A 300 mm diameter wafer would have millimeter-level lateral resolution as compared to profi le based resolution given by existing techniques. This greater data density provides extremely valuable information to the quality of the wafers. We will compare and contrast different metrology techniques and their relative attributes and discuss additional developments in using this technology. The signifi cant advantages provided by this approach for precision characterization of wafers and wafer stacks will also be provided. Background In the beginning the semiconductor industry was just an emerging market. Lithography as it is done today was beyond the imagination of even the people at the leading edge of this new technology revolution. Wafers were small, fi rst 1 inch then 2 inches in diameter. The industry was looking for consistent characterization of these small thin wafers to establish standard quality. What was it they wanted to characterize? They were looking for a measure of the degree of convex/concave shape in the wafer, and an overall wafer fl atness measurement. With the wafer sizes of the time, it was desirable to support the wafer in a simple manner that was easily reproducible, so the three-point mount was perfect. It is a kinematic support, so any three-point support should result in the same defl ections. A small misalignment of the part would result in a relatively small reproducibility error. However, measuring the concave/convex magnitude becomes trivial in this fi xture. You can simply measure the center point and compare the measurement to the measurement of an optical fl at supported by the same three points. This then becomes a measure of the sag of the wafer. There is a small amount of gravitational infl uence, but this should remain relatively constant from wafer to wafer for the same nominal geometry. The beauty of this measurement is that you can measure bow with a single point probe on a fi xed jig. Film stress could be correlated directly to the magnitude of the change in this bow measurement (e.g. see Stoney’s equation) after the application of the fi lm. 2 Measuring warp still requires full surface information, but the three-point support allows the measurement to be made directly without the complication of calculating the least squares plane, making it convenient back when these types of analysis were limiting factors. Over time the diameter of the wafer grew, but the thickness did not increase proportionally, 3 inch, then 4 inch, 5 inch for a while, then 6 inch, 8 inch, and now 300 mm (12 inch) with 450 mm (18 inch) just over the horizon. This seems like no major concern, but all those small errors associated with minor alignment errors start to become very signifi cant relative to the target values of bow/warp. Another challenge arises from variation from how the wafer is mounted on the metrology tool. Often times a threepoint mount is used with characterizing a wafer, but fourpoint mounts, ring supports and others are also utilized. Deformation from gravity will signifi cantly differ in shape and magnitude depending on how the wafer is held during characterization. Figure 1 shows the shape of a theoretically perfect (TTV and fl atness = 0 μm) wafer if it is held at the perimeter by a three-point (1a), four-point (1b), or ring support (1c). The magnitude of the total defl ection (sag) is also strongly related to the mounting strategy. As indicated in Figure 1, the calculated defl ection through fi nite element analysis (FEA) modeling of a 300 mm diameter, 0.7 mm thick supported at the perimeter by three-point, four-point and ring support is 206 um, 160 um and 130 um respectively. Figure 1. FEA showing shape of deformation with different support levels The effect of how the wafer is supported on the total sag of the wafer was discussed above. There can also be substantial changes in variation by small changes in the wafer properties, mounted under the same conditions. Let’s consider a few simple cases: Wafer diameter: 300 +/1 mm Wafer material: Silicon Density: 2.33 g/cm3, Elastic modulus: 141 GPa Poisson’s ratio: 0.22 Wafer thickness: 0.7 mm +/0.01 mm Three-point support radius: 147 mm Glass Material: Corning SGW3 Density: 2.38 g/cm3, Elastic modulus: 74 GPa, Poisson’s ratio: 0.23 Wafer thickness: 0.7 mm +/0.01 mm Three-point support radius: 147 mm The fi rst thing to note is that for these material properties, the magnitude of the gravitational sag from a three-point support at -3 mm from the edge is 206 microns, which is certainly not negligible. Compare this to the results for a 0.4 mm thick, 50 mm diameter Si wafer, which has a sag of just over 0.35 microns. This, you can argue, is negligible, the variation from loading is almost certainly negligible, and the variation from different wafer thickness is also negligible. If you consider our 300 mm wafer case, simply varying the thickness of the wafer from 0.69 mm to 0.71 mm changes the gravitational infl uence by over 12 microns. Clearly for getting a meaningful measurement with a threepoint support requires compensating for the infl uence of gravity. However as we can see from the sensitivity to the wafer thickness, the compensation is highly sensitive to variations from wafer to wafer. Even with constant wafer geometry and properties, measuring a wafer with this kind of magnitude from gravity becomes unnecessarily complex, and incredibly sensitive to load orientation. For wafers with relatively loose tolerances, 10 μm TTV and 200 μm warp for example, this may appropriate. However, gravity compensation is a questionable strategy to obtain accurate measurements for 300 mm wafers with 2 μm or 3 μm of TTV and 40 μm or 50 μm of warp. Efforts underway to increase the diameter of the semiconductor wafers will exacerbate the issue. For the purpose of illustration, consider the same 0.7 mm thick wafer, with a 450 mm diameter. This will sag more than 1000 um. Different materials such as glass, typically have stiffness lower than silicon and the infl uence of gravity becomes even greater. For example, a glass wafer with the same geometry as our 300 mm silicon wafer will sag by 404 microns instead of the 206 microns described previously. Table 1 summarizes the sag as a function of wafer diameter, thickness and material. With larger and thinner wafers, a three-point support is likely to introduce as much uncertainty in the measurement as the magnitudes of the real wafer fl atness. Using other non-kinematic support methods will not allow for accurate compensation either. Table 1. Summary of wafer sag variability with wafer dimension/material (1a) Three-point mount at perimeter Sag: 206 μm (1b) Four-point mount at perimeter Sag: 160 μm (1c) Ring support at perimeter Sag: 130 μm Material Diameter (mm) Thickness (mm) Support Radius (mm) Sag (μm) Silicon 50 0.40 22 0.35 Silicon 300 0.69 147 212 Silicon 300 0.70 147 206 Silicon 300 0.71 147 200 Silicon 450 0.70 222 1060 SGW3 300 0.70 147 404