Documentos de Académico
Documentos de Profesional
Documentos de Cultura
Master of Science
January 1997
Abstract
Reliable assessment of the dynamic behavior and mechanical properties of a silicon wafer is critical for the minimization of rattling and walking during manufacturing of semiconductors. This paper presents the results of a thorough study of a wafers response to vibrations. Finite element analysis and several experimental tests were performed to verify that the walking/rattling problem was properly modeled and the frequencies questioned were correct. With an accurate dynamic understanding of the wafers, the behavior of larger wafers can be predicted as wafer technologies develop.
Acknowledgments
I am eternally grateful to Dr. Eric Marsh, my friend and research advisor, who saw my potential and gave me the support, guidance, and advice needed to accomplish this goal. He also acted as a Western Pennsylvaniaism translator and showed me the way when things needed fixed. I would like to thank Motorola MOS 12 of Chandler, Arizona for funding this project and supplying such a diverse problem to learn from. I would especially like to thank Carl White who served as a contact and technical advisor. I would like to thank my mother, Linda Maher, for giving me constant encouragement and for supporting all of my decisions. I would like to thank my father and true mentor, Thomas Maher Sr., whose almost limitless mechanical knowledge and abilities never cease to amaze me. Finally, I would like to thank my grandfather, Joseph Giebel, whose life as a mechanical engineer has been an inspiration. He is missed immensely.
1. Introduction
Many semiconductor process tools are known to exhibit undesirable structural vibrations. These vibrations can cause rattling and walking of the silicon wafers in the cassettes. As a result, the wafers can be damaged by particle generation or by fracture. These vibrations may also cause the wafer to slide from their original position. This precarious position leaves the wafer in the path of robotic arms and other handling devices. If a wafer were shattered by a process tool, the damage would not be limited to that wafer only. The airborne fragments of silicon may also contaminate other wafers. Causes of problematic external vibrations include translational and rotary axis motion, robotic automation, vacuum pumps, or vibrations from surrounding process tools. Many studies had been done to maximize production rates and to solve problems associated with high output production. Others have investigated quality control problems but no study found to date studies dynamic problems of the wafer. The Machine Dynamic Research Lab's (MDRL) purpose in this study was to provide an understanding of the rattling problem so it can be reduced or eliminated. As wafer sizes increase from the current 200 mm to 300 mm and more, it is expected that these vibration issues will become increasingly critical.
To further understand the motion of the silicon wafer, experimental modal analysis was also used to observe deflection shapes at particular frequencies. Due to the nature of the object being tested, special considerations and testing procedures where needed for an accurate study. Finally, high speed video during machine excitation was exhaustively studied. All three procedures will be detailed in this paper.
The cassette can be excited in any direction as a result of external vibration. To understand the dynamic problem, separation of X, Y, and Z motions was desired to conduct an accurate study. Movement for this study included all three coordinate axes. The coordinates are shown in Figure 2.1 and will remain constant throughout the study. Testing devices were developed to individually excite the cassette in the horizontal (X and Y axes) along with the vertical direction (Z axis).
z
X
Figure 2.1 Coordinate axes with reference to wafer position. Stability test fixtures also had to allow different cassette configurations (e.g., full cassette, single rail, etc.). Figure 2.2 shows three cassette styles used to observe stability. Both a single slot and a single slot with the top rail removed (half slot) were cut from a full cassette. These modifications were made to study the effects of boundary conditions. The full cassette was also studied to determine the movement of whole structure. This would give an indication if the dynamics of the cassette effect the wafer's stability. For quick reference, the three cassette styles are listed and described below: Full Cassette: Single Slot: Half Slot: Single wafer in top slot (#25) of complete cassette. Single wafer in a single cassette slot which was cut from a complete cassette. Single wafer on a single cassette slot with no top guide which was cut from a complete cassette.
Full Cassette
Single Slot
Half Slot
Figure 2.2: Cassette configurations for stability testing Two different cassette materials were also used to study the effects of material. Black and white cassettes with two distinct stiffness differences were loaded with 25 wafers and tested. The black cassette had a much higher density and more structural stability. This showed that cassette dynamics had a minimal effect on wafer stability.
OF SD otion M
Figure 2.3 Horizontal testing apparatus. The first natural frequency of the test fixture was considered to prevent resonances from occurring in the test frequency range. The testing frequency range needed to be below the first resonance frequency of the test apparatuses to minimize dynamic interference. To calculate this, the stiffness of the structure must first be determined. This is easily determined by beam theory that is shown by Equation 2.1. It was determined that the mass of the table plate and test specimen would not exceed 3 kg and that apparatus would be designed for a first natural frequency above 100 Hz. The required stiffness of the structures was then calculated to be 9.6E5 N/m using Equation 2.2. 2 Ewt 3 = L3 where: E = I = w = t = L =
k eq
[2.1]
modulus of elasticity moment of inertia width of beam thickness of beam length of beam
The undamped equation for a simple degree of freedom mass/spring model shows that the first natural frequency relates to stiffness and mass of the equivalent system. See equation 2.2.
k m
[2.2]
where: = k = m =
The final design had a first natural frequency at approximately 90 Hz. This was desirable because it was higher than the test range yet flexible enough to have sufficient motion for low force levels. The base of the test apparatus was securely fastened to a 1 ton machine base assuring proper grounding. The flexure table was excited using a 90 N Ling electromagnetic shaker model V4. The excitation level and frequency to the shaker were supplied by a digital signal analyzer (HP 35670A). The analyzer had a control loop function which constantly measured and observed the flexure table vibration with a non-contacting capacitance probe.
SDOF Motion
Figure 2.4 Proposed vertical test apparatus. Instead, a beam constrained on both ends was used to replace the flexure table in the vertical direction. The dynamics of the beam were predicted using beam theory in the same way used for the horizontal apparatus. For a beam with clamped boundary condition the first natural frequency is calculated from Equation 2.3. Using an aluminum beam 15 cm wide, 51 cm long, and 6.35 mm thick the natural frequency was 127.8 Hz.
2
4.73 = L
EI A
[2.3]
where: E = I = w = A = L =
modulus of elasticity moment of inertia width of beam cross sectional area length of beam
The beam width and length were used to limit unwanted motion in the off-axis directions. A thin beam was used to allow sufficient motion in the desired vertical direction. With these dimensions, the maximum static displacement was calculated using beam theory as shown in Equation 2.4.
PL3 = 48 EI
where: P = L = E = I =
[2.4]
With a 90 N force, the max. amplitude of the vertical apparatus was calculated to be 2 mm which was sufficient for stability testing. Again, the fixture was mounted to a machine base to closely resemble clamped boundary conditions used by the model. Also the same control loop and shaker configuration was used to excite the structure. This vertical apparatus is shown in Figure 2.5.
SDOF Motion
Shaker
2.5. Experimental Modal Analysis 2.5.1. Experimental Modal Analysis of Silicon Wafer
Mode shapes, natural frequencies, and damping factors can be determined experimentally using techniques such as experimental modal analysis (EMA). When a measured force is applied to the structure, transducers are used to measure the response. A dynamic signal analyzer uses this information to estimate a frequency response function (FRF). The FRF is a complex quantity that varies as a function of frequency. The output response is measured using a transducer such as an accelerometer, velocimeter, strain gage, or a displacement probe. Common excitation methods include impacting the structure with a hammer or by driving the structure with an electromagnetic shaker. The input force is measured using a force transducer. Once FRFs are calculated over many points on the structure, they are then analyzed to determine the mode shapes, natural frequencies, and damping factors.
For a structure such as a silicon wafer where the overall weight of the object is very low, mass loading for the transducer has a great effect on the results. A large transducer increases the mass of the system causing the natural frequencies to drop. This makes testing very difficult and the results unreliable. Fortunately, a non-contacting capacitance probe may be used over a wide range of frequencies. They are very accurate but function over a very small measuring range. Since this is the case, impacting is an unacceptable means of excitation because the wafer would move out of the range of the probe after impact. Exciting the structure with a shaker would also be unacceptable because a load cell would cause mass loading. The only non-contacting means of exciting a structure is by acoustic wave propagation. Figure 2.6 shows the setup used for this testing. A speaker is placed above the wafer and a pure sine wave is output from the speaker. The response is measured with a Lion Precison capacitance probe model #DMT10 for the points on a grid pattern drawn on the wafers surface. A grid pattern was selected which was small enough to correctly represent the mode shapes. This sine sweep was done for a range from 20 to 320 Hz.
Figure 2.5 Setup for acoustical exciting technique with free boundary conditions. The wafer
is resting on foam with a speaker and cap probe in place. By definition, this is not a pure modal analysis because values of input amplitude are not reliably captured. However, mode shapes are obtained because the input force is constant throughout the testing.
The wafer is placed on packing foam to allow freedom of movement. Ideally, free boundary conditions are desired. The foam support does not provide perfect free boundaries because the wafer bounces on the foam at non-zero frequencies. However, the foam does provide a reliable approximation. Testing was performed twice, one using the free conditions as explained above and a second time using the boundary conditions of an actual cassette. The same acoustic excitation technique was used but the wafer was supported by the single slot rather than packing foam. This was desired to better represent actual conditions of the wafer during processing.
experimental setup was also constructed so that the theoretical deflection could be compared to the experimental results. The modulus of elasticity of the FEA model was iteratively changed until the experimental deflection matched the FEA results. The setup is shown in Figure 2.7 below. This fixture used three points of contact in an equilateral triangle configuration to ensure that wafers loading was distributed evenly. Point contacts were made by using hardened high carbon steel ball bearings mounted to the top of a base. This three point kinematic coupling configuration gives good stability and overall stiffness in the horizontal plane. The three balls also provide easily modeled boundary conditions for the FEA model.
Force
The FEA model of the silicon wafer has the same boundary conditions and as the actual static configuration. The finite element model is show in Figure 2.8.
Force
Figure 2.8 Finite element model with boundary conditions to verify Youngs modulus.
Even this simple experiment required some preliminary convergence checking of the FEA model. All finite element models must be solved several times with varying amounts of elements (mesh density) to ensure that the solution converges to an unchanging answer as the number of elements increases. This convergence checking is necessary because finite element models with insufficient mesh density will not provide reliable results. A mesh with too few elements will overestimate the stiffness of the structure. A converged model gives confidence that the model accurately represents the stiffness of the wafer. The different model sizes are plotted in Figure 3 to show the convergence of the wafer model with increasing mesh density (number of elements).
8.8
x 10
-5
8.6
8.4
8.2
7.8
100
200
500
600
700
Since the static FEA acceptably represents the real structure, a dynamic analysis can be solved with confidence. Material constants used in this study for monocrystalline silicon are show in the table below.
Table of Material Properties for Monocrystalline Silicon
Motorola Supplied
Density (g/cm3) Young's Modulus E (dynes/cm2) 2.328 1.884 e12
Calculated
2.328 1.47e12
constraints. For the wafer being placed being within the cassette, the FEA model was restrained in the Z direction at opposite ends of the wafer, as shown in Figure 2.9.
Figure 2.9 Wafer with vertical constraints simulating its contact with the cassette.
The results show that boundary conditions are very relevant and wafer constraints play an important role in dynamic frequencies.
3.2. Experimental Results 3.2.1. EMA of Wafer with Free Boundary Conditions
Frequency response functions from experimental findings were curve fit and reconstructed using a commercial software package. The first and second natural frequencies of the wafer with free boundary conditions were 186 and 240 Hz. The reconstructed mode shapes are shown in Figure 3.3.
# 8:239.75 Hz
# 7:185.75 Hz
Figure 3.3 EMA results of the wafer with free boundary conditions.
The first mode shows bending at 186 Hz. The second mode shows a diaphragm shape at 240 Hz.
# 1:49.88 Hz
# 2:173.88 Hz
Figure 3.4 EMA results of the wafer with constrained boundary conditions.
The 50 Hz mode shows bending across the width of the wafer. The mode at 174 Hz shows a twisting motion.
# 1:66.73 Hz
# 2:93.72 Hz
# 3:215.51 Hz
# 4:263.11 Hz
Modes 1 and 2 are dominated by a shearing motion in the cassettes. Modes 3 and 4 show buckling of the cassette walls.
3.3. Analytical Results 3.3.1. FEA of Wafer with Free Boundary Conditions
All FEA results were calculated using a commercial finite element package. The first two natural frequencies and corresponding mode shapes obtained from analytical analysis are shown for a wafer with free boundary conditions in the Figures 3.6 and 3.7. These critical frequencies have been shown to be 152 Hz and 244. The outputs from the FEA shows these as modes 1 and 2.
Figure 3.6 First natural frequencys mode shape as calculated by FEA. (152 Hz)
Figure 3.7 Second natural frequencys mode shape as calculated by FEA. (244 Hz)
Figure 3.8 First natural frequencys mode shape as calculated by FEA. (54Hz)
Figure 3.9 Second natural frequencys mode shape as calculated by FEA. (152Hz)
4. Discussion
In this section, the results of the experimental and analytical testing are reviewed to provide an overview of the wafer study. The results suggest that the wafer walking problem is most severe as a result of machine vibration in the vertical (Z) direction. However, wafer walking was observed even when the excitation occurs in the X direction.
cassette test with X direction excitation. Low rigidity of the cassette allows this small swaying motion to occur.
before technological changes are implemented. Cassette geometries and contact point configurations could also be investigated using finite element analysis to optimize wafer stability during processing. Finite element analysis shows that the next generation wafer (300 mm) results in the first natural frequency being 24 Hz. With the excitation levels of current semiconductor process machines, walking/rattling may happen more frequently in 300mm wafers. To safely manufacture 300 mm wafers, several issues should be considered to limit the walking problem. First, different contact points for the wafer could be investigated. The current configuration uses two point contacts on the edges of the wafer. For maximum stability, a kinematic three point support is suggested. This third point would not only increase stability but effectively increase the stiffness to the wafer thus increasing its modal frequencies. Second, a cassette material with better damping characteristics should be investigated. This would decrease the excitation transmitted to the wafer and increase its stability threshold. Current wafer cassettes are constructed from a hard plastic which offers little damping. Finally, threshold limits to the cassette should be observed during the automation process. Machine vibration amplitudes should at no time exceed the threshold limits shown to cause walking for its corresponding frequency and orientation. To minimize extraneous vibrations, vibrational isolators should be used on vacuum pumps, process tools, and transporting devices. To limit vibrations during automation, machines could be monitored by accelerometers to measure peak values. If the threshold limit is violated, the processing and all surrounding vibrations should be stopped. The wafers motion should then be let to settle before processing is continued. In future studies, wafer stresses caused by dynamic motions need to be investigated. The dynamic bending motion results in high surface stresses which may cause problems in proper etching. This is of large concern in the 300 mm and larger wafers because of its low stiffness and low threshold limits.
References
1. 2. 3. 4. 5. 6. 7. Avallone, A.A., and Baumeister, T., Marks Standard Handbook, McGraw-Hill, New York, 1995. Hodgman, C. D., Handbook of Chemistry and Physics, Chemical Rubber Publishing, Ohio, 1939. Meirovitch, L., Analytical Methods in Vibrations, Macmillan Publishing, New York, 1967. Pytel, A., and Singer, F.L., Strength of Materials, HarperCollins Publishing, New York, 1987. Rao, S.S., Mechanical Vibrations, Addison-Wesley Publishing, New Jersey, 1990. Structural Dynamics Research Corp., I-DEAS Master Series Student Guide, Ohio, 1994. Trethewey, M.W., ME554 - Experimental Modal Analysis Lecture Notes, Department of Mechanical Engineering, The Pennsylvania State University,1996.