A new type of scattered light measurement method will be presented, capable of measuring the full wafer surface of a 300 mm wafer in less than 30 seconds. Besides the roughness, the sensor simultaneously measures warpage, waviness and defects.
BY R. BRODMANN, B. BRODMANN, K. KONOVALENKO, and C. WIEHR, OptoSurf GmbH CHING-HSIEN HUANG, YA-LENG CHEN, Amkor Technology, Inc.
The trend towards very small and high-density electronics requires advanced processes to meet the specifications of thickness and thermal properties of the devices. This means that the processed silicon wafers have to be thinned from their original thickness of more than 700 μm down to 50 μm or less. The most common and relative low cost thinning method is back grinding by means of mechanical removal of the residual silicon. The wafer is fixed on a porous vacuum chuck with the IC (integrated circuit) side down. The rotation axis of the grinding wheel is positioned off-axis to the rotation axis of the wafer (distance is the radius of the wafer). The chuck has a slightly conical shape which deforms the wafer with a very little tilt to ensure that the grinding wheel only contacts half of the wafer during the grinding process. Due to the rotation of the chuck and simultaneously rotation of the grinding wheel a typical spiral pattern of scratches on the wafer surface is generated.
Depending on the grit size of the grinding wheel and the machining parameters as rotation speed and feed rate, this mechanical impact is responsible for the roughness, stress and induced subsurface damage. Therefore, a modern wafer grinding machine begins with a coarse grinding wheel to get a fast removal of the silicon and at the end follows a fine grinding process step with small grit size grinding wheel. This final process is absolute necessary when thinning down to 50 μm in order to minimize subsurface damage and stress. The roughness of the surface should be often in the range of Ra <10 nm or even 1 nm which is a challenge for mechanical grinding machines. Is the roughness too high or not uniformly distributed on the wafer surface, the later process steps as wire bonding, flip chip assembling, molding and testing can damage the thin chip through breakage. Besides a low surface roughness, the fracture strength of the die after dicing also depends on the orientation of the grinding marks. The correlation of die strength with roughness and surface texture is described in Ref. 1 and 2.
The interaction of the grinding wheel with its large number of single cutting edges, undergoing non-uniform wear, and the silicon surface, in particular when applying the fine grinding procedure is a rather complex process. Therefore, it is not possible to predict the quality of the entire wafer surface after grinding by means of a few small area roughness measurements with an AFM, a WLI or CFM, which is the standard today. Typically, the assessed area of one single measurement is 20 μm x 20 μm in case of an AFM and 160 μm x 160 μm with a CFM or WLI. Each measurement takes about 20 s-30 s and requires anti-vibration equipment to avoid influence from environmental mechanical noise.
In order to get information of the entire wafer surface, much faster and more robust measurement techniques are necessary. Scattered light measurement is the only method which can achieve these requirements. In the present paper, results of a new measurement machine (WaferMaster 300) are discussed which uses a scattered light sensor  to measure the roughness of a full 300 mm wafer surface in less than 30 s. Due to a special design of the sensor the WaferMaster can measure in addition the warpage, waviness and defects.
Measurement principle and surface characterization
Using scattered light to measure surface defects and roughness is already well known for CMP polished bare wafers, pattered wafer, hard disks, mirror surfaces and high quality fine machined automotive parts. The new type of scattered light sensor to measure back- grinding wafer is shown in FIGURE 1.
The light source (1) illuminates nearly perpendicular the wafer surface with a 670 nm red LED spot of 0.9 mm spot size (2). This is the standard modus for fast measurement with medium lateral resolution. For high lateral resolution, another spot with 0.03 mm diameter from a laser source with the same wavelength can be switched on. The optics (3) collect the scattered light in an angle range of 32° and guides it to the linear detector (4). In contrast to other scattered light sensors this sensor measures the specular light (0°-part of the surface) together with the scattered light created by the microstructures of the surface. The advantage of this set-up is the capability to use the center of gravity of the scattered light distribution (5) as signal of the local geometrical deformation of the surface. Knowing the local slope angle of the surface and measuring continuously the surface in equal distance (created by an encoder signal) the local height can be calculated and, by integration of all angles, the entire profile of the surface.
The chuck with the wafer (6) rotates continuously during the measurement and the sensor moves linearly from the wafer edge to the center. Subsequently the sensor measures the entire wafer surface and assesses on a 300 mm wafer in the standard modus (0.9 mm spot) about 60.000 single roughness measurements in 30 s. Very important is an additional rotation (7) of the sensor because the linear detector should be always orien- tated normal to the grinding marks to get the maximum roughness value. As roughness parameter, the variance of the scattered light distribution Aq is calculated (FIGURE 2). ψi are the single scattered angles, M is the center of gravity and p(ψ) is the distribution curve.
The advantage of Aq is the close relation with the profile slope parameter Rdq which describes surface friction very well. To follow the Semi standards in which the mean roughness Ra is established as roughness value, the Aq parameter was correlated with Ra by comparison measure- ments of different wafers with a confocal micro- scope. Due to the stochastically property of the amplitude distribution of the ground surface there is a rather good correlation between Aq and Ra even when using different grit size of the grinding wheel. But it should be taken into account that Aq is a more versatile parameter, because it reacts on both the vertical and lateral structures of a profile whereas Ra only measures the mean vertical height. This property of Aq could be interesting for characterizing die strength and should be investigated in more details in the future. In FIGURE 3 the measured correlation is shown. Several wafers were in- vestigated on different areas and ground with different grit sizes from #2000 to #8000. In addition, a CMP polished wafer with a Ra value <1nm was measured to check the accuracy of the system. In order to calculate the Ra-value the fitted correlation equation is used in the WaferMaster machine.
As already mentioned, the scattered light sensor has a second evaluation channel to measure warpage and waviness by means of slope angle analysis. As shown in Fig. 2,the measurement beam is deflected under 2x the local slope angle θ. Therefore, the scattered light distribution is shifted on the linear detector by the angle value M. θ can be measured by using the first statistical moment of the scattered light distribution curve. Knowing the step size ∆x from an encoder and the focal length of the optics, the local height ∆y can be calculated and by sum up, the height profile can be generated.
In FIGURE 4 the roughness results of 3 wafers are shown, each 300 mm size. They all were ground with the same grinding wheel (grit size #4000), but using different grinding machines. In total 40.000 measurements were taken in 25s with the 0.9mm spot. Besidesthe difference in the mean roughness value, it demonstrates in particular that the machines did leave its own characteristic pattern. The interpretation might be interesting to analyze in detail the grinding parameters as feed rate, chuck geometry, rotation speed, and others.
An example of the simultaneous measurement of roughness, form (warp) and waviness of a 200 mm back grinding wafer can be seen in FIGURE 5. Although the grinding wheel was also grit size #4000, the mean Ra value is a bit higher. From the grinding marks pattern, it can be seen that the rotation was counterclockwise which changes the orientation from left to right. The warp is rather high because no vacuum was used. The waviness was calculated by applying a 50 waves high-pass filter. The filter is working on the circumference which means that the center area is filtered strongly than at the edge and middle area. Different filter method will be used in the future. The waviness structure follows the roughness pattern, but there are also visible some superimposed weak linear stripes from left to right. These stripes are more prominent in the following measurement (FIGURE 6), which is the result of another 200 mm wafer but ground with a #2000 grit wheel. The interesting point is not the higher roughness, which is induced by the coarser grinding wheel, but that the stripes here are more prominent than the waviness pattern of the grinding marks. The peak to valley height evaluated from A to B is more than 1 μm, which is about 10 times the profile height of the grinding marks waviness.
The linear stripes are probably caused by the previous wire sawing process, which did not vanish after the grinding process. This could happen, because the wafer is fixed by vacuum on the chuck during grinding which makes the surface temporarily flat. When the wafer is released after the grinding process the waviness structures return. This phenomenon is investigated and described by Pei et al . Furthermore, if the chuck is not cleaned very well the same characteristic can create bumps and dimples. An example of dimples is shown in FIGURE 7. The waviness map of a 300 mm polished wafer is covered with 2 larger and some smaller dimples. By using the 0.03 mm sensor spot the larger dimples where measured again with higher local resolution and represented in a 3D map. The width is in the mm range whereas the depth does not exceed 1 μm.
Another example of high resolution measurement can be seen in Fig. 8. These measurements were done with an x/y-scanning module covering an area of 40 mm x 20 mm, also by using the small spot size of 0.03 mm. The measurements represent the waviness structures (after applying a 25 waves high pass filter). The mean roughness is 5 nm Ra. Near the center of this section another dimple is visible. The selected profile (a) shows the general waviness with a peak to valley height of about 30 nm. Repeatability measurements have shown that structures of 1 nm height could be resolved.
This makes the WaferMaster moreover interesting for the assessment of nanotopography structures to measure the planarization quality after CMP processes. Also, as can be seen in the 3D graphic, the small spot is able to detect single defects (red peak at the right side) and it has to be investigated, what the limit of lowest defects is. Certainly, it cannot compete with the much more powerful scattered light systems, especially designed for small defect detection in the front-end industry, but it is sufficient to use this function in backend processes.
A new scattered light sensor technique was presented to measure wafer surfaces, particular in the field of back grinding. The sensor combines surface roughness measurement by means of evaluating the variance (Aq) of the scattered light distribution and use additionally the method of deflectometry to assess form (warpage) and waviness. The Ra evaluation is based on correlation measurements with a confocal microscope. It could be shown that the sensitivity of roughness measurements is going down to Ra = 1 nm with an accuracy of 0.1 nm. The advantage of this technique is the speed (25 s for a whole 300 mm wafer scan) and the ruggedness against environmental mechanical noise. The capability of the full area representation of roughness, warpage and waviness opens new possibilities to characterize and improve the grinding processes as well as checking the quality from the edge area to the center completely.
Depending on the packaging design and the sensitivity of the processes which follow after the back grinding, the difference of the roughness from edge to the center and along the circumference, as well as strong warpage, waviness and defects can influence the final function and performance of the singulated chips. Die breakage e.g. directly depends on the roughness and in particular on the grinding marks orientation. Therefore, a fast and continuous measurement of the back-grinding quality can help to improve the yield in the backend process.
We would like to give special thanks to Kevin Hsu from Sanpany and Ian Chen, Honjang Global Technology for their kindly support in organizing the wafer samples and to confirm our CFM measurements with an WLI microscope.
 Michael Raj Marks, Zainuriah Hassan, Kuan Yew Cheong, Ultrathin Wafer Pre-Assembly and Assembly Process Technologies; Critical Reviews in Solid State and Materials Sciences, 40:251–290, 2015, DOI: 10.1080/10408436.2014.992585
 Desmond Y.R. Chong, W.E. Lee, B.K. Lim, John H.L. Pang, T.H. Low, Mechanical characterization in failure, strength of silicon dice, 2004 Inter Society Conference on Thermal Phenomena, 2004 IEEE
 Z.J. Pei, Graham R. Fisher, J.Liu, Grinding of Silicon Wafers: A review from historical perspectives, international Journal of Machine Tools & Manufacture, 48 (2008) 1297-1307
 Seewig, J., Beichert, G., Brodmann, R., Bodschwinna, H., and Wendel, M. 2009. Extraction of shape and roughness using scattering light. In Proceedings of SPIE. Optical metrology, Systems for Industrial Inspection VI 7389.