13 (2003) 686–692


A measurement of Young’s modulus and residual stress in MEMS bridges using a surface profiler
M W Denhoff
Institute for Microstructural Sciences, National Research Council, Ottawa, Canada K1A 0R6 E-mail:

Received 18 November 2002, in final form 2 May 2003 Published 20 June 2003 Online at Abstract This paper addresses a relatively simple method of measuring Young’s modulus and residual stress in microelectromechanical system (MEMS) type structures. A surface profilometer is used to measure the deflection of thin film fixed–fixed beams due to the force applied by the profilometer probe. These measurements are analyzed using analytical beam theory. The treatment of end effects and the accuracy of the measurement are discussed. Measurements and results are presented for PECVD grown silicon nitride films.

1. Introduction
The mechanical properties of the materials used in microelectromechanical system (MEMS) fabrication are of fundamental importance. This knowledge is needed for the design of the devices, and measurement of the properties is needed to check the consistency of materials during fabrication [1]. Since the material properties are sensitive to small changes in process, it is desirable to be able to determine material properties across a wafer and to determine wafer to wafer variation. A number of methods have been previously reported. Some of these require complex, expensive equipment and complex data analysis. Others are simple but not too reliable. A relatively simple and inexpensive method would be useful. Surface profilers are found in most micro fabrication laboratories and it would be good to be able to use this existing equipment. There have been some previous reports describing the use of a surface profiler to measure the deflection of MEMS fixed–fixed beams and determine Young’s modulus from this. Some reported methods rely on either simple (and incomplete) mathematical models or finite element models of the MEMS beam as well as depend on a separate determination of the residual stress in the beam [2, 3]. The simple models omit tension due to stretching of the beam, but this is important in the fixed–fixed case. Zhang reports a detailed and complex analysis of the problem, using a combination of analytical and finite element computation to determine end effects, as well as using a nanoindenter instead of a surface profiler [4].

There is other published work which uses interferometry to measure beam displacement and a complex numerical model to determine mechanical properties of MEMS structures [5]. This paper presents a middle ground using the analytical mathematical theory. The analysis of the mathematical model is extended so that both Young’s modulus and the residual stress can be determined from surface profiler deflection measurement. A mathematical model for the deflection of a fixed–fixed beam under the application of a force at an arbitrary position along the length of the beam will be presented in the first section. Then three methods of analyzing surface profiler measurements will be developed. Results will be shown for silicon nitride (SiN) films grown by PECVD. Finally, shortcomings of the method will be discussed, in particular, inaccuracies using the profiler probe and modeling the ends of the beams.

2. Theory
Consider a rectangular beam, clamped on both ends with a force, W , applied a distance, Q, from the left end. It will deflect as shown in figure 1 (case 1). From basic elasticity theory [6], for the case where the deflections are small compared to the length of the beam, the deflection of the beam (in the section to the left of the point where the force is applied) must satisfy the differential equation, −W (L − Q) . (1) EIy (x) − T y (x) = L 686

© 2003 IOP Publishing Ltd Printed in the UK

Ts . W (L+Q) T L cosh k(L − Q) (9) Equating the two above equations and solving for D1 . For the section of the beam to the right of the point where the force is applied. I = wt 3 /12. This condition gives D2 = − W + T + D1 = W Q T L W (L−Q) T L cosh kQ − D1 k sinh kQ [k sinh k(L − Q)]. Putting in the full expressions and rearranging the terms gives y2 (x) = D2 = − W T + W (Q) T Lk (L−Q) Lk y(x) is the displacement of the beam. When Q = L/2 both of these reduce to D1. the change in length is well approximated by 1 L 2 L= y dx. the integral in equation (4) must be performed. It is important to understand that the force due to tension is composed of two parts. The length of the deflected beam is found using the integration formula L+ L= 0 L sinh kQ + D1 cosh kQ − D1 [cosh k(L − Q) − 1]. (4) 2L 0 Now the solution. L 2T k sinh k 2 (11) For slight bending. One part. However. A beam being deflected by a force. a distance R from the left end of the beam. The other part. Since the two cases are reflections of each other.Young’s modulus and residual stress x case 1 Q W R y L x W R y case 2 F1 = − D 1 and E1 = − W (L − Q) C =− . which gives D2 = W Q (cosh kL T L − cosh k(L − Q)) + W (2L−Q) T L 1 + y 2 dx. such that R = L − Q. In order to find the numerical value of T. An applied force W generates shearing force W (L − Q)/L. Tr . If the beam was under compression. This is case 2 in figure 1. T L EI The boundary conditions at the clamped end are y(0) = 0 and y (0) = 0 which give All the constants have been determined and equations (6) and (7) describe the complete solution. (5) Substituting this back into equation (1) gives W (L − Q) T C1 = and k2 = . T is the axial internal force on the area of the beam due to tension. k T Lk The solution can now be written as W (L − Q) W (L − Q) x− y1 (x) = T L T Lk × sinh kx + D1 cosh kx − D1 . The equation describing the second case can be obtained by the substitution of Q → R = L − Q which gives W (Q) W (Q) x− sinh kx + D2 cosh kx − D2 .2 = −1 W cosh k L 2 . (8) sinh k(L − Q) The slope will be continuous at the point on the beam where the force is applied (unless the stiffness equals 0 as for a string). is due to the stretching of the beam as it is deflected. the − sign in front of T would be changed to a + sign. which is constant in the left section of the beam. While a constant as a function of x. W (L−Q) (cosh kL T L − cosh kQ) + × (1 − cosh k(L − Q)) [k(sinh kL − sinh kQ − sinh k(L − Q))]. the slopes y1 (Q) = −y2 (R). its value depends on the final length of the deflected beam. The result is L 0 y 2 dx = 2 C2 C1 sinh kQ cosh kQ + 1 Q 2k 2 2 C 2 + C1 Q − 2 1 sinh kQ + 2C1 D1 cosh kQ − 2C1 D1 k D2k − C1 D1 cosh2 (kQ) + C1 D1 + 1 sinh kQ cosh kQ 2 2 2 2 C2 D1 k Q+ sinh k(L − Q) cosh k(L − Q) − 2 2k 687 . consider the beam with the force applied at a new point. the shearing force is W (L − R)/L. Rather than directly work with the solution for the right side of the beam. y(x). (6) Figure 1. This is a straightforward but lengthy calculation. 2 0 Then the force of tension due to stretching is Ts = (3) EA L 2 y dx. a prime represents differentiation with respect to x and the origin is chosen to be at the left end of the beam. (2) × (1 − cosh kQ) [k(sinh kL − sinh kQ − sinh k(L − Q))]. Case II is a mirror image of case I. of equation (1) must be found for the appropriate boundary conditions. the tension is only implicitly defined. The line drawn to represent the beam is a solution of equation (1). where w is the width of the beam and t is its thickness. E is the Young’s modulus and I is the moment of inertia of the cross-section of the beam. The general solution for the left section of the beam is y1 (x) = C1 x + E1 sinh kx + D1 cosh kx + F1 . is due to the residual stress in the thin film and is a constant. T = Tr + Ts . Continuity of the beam at the point where the force is applied will be used to find the last constant. (10) Now the coefficient D2 can be found from D1 by substituting Q → R = L − Q. (7) T L T Lk Continuity requires that y1 (Q) = y2 (R).

must be solved and then the value for the deflection. for the stretching tension. Leff . 3. The beam is suspended above the substrate. One was a Dektak 3 which had a constant probe force2 . Image of the Dektak probe about to scan along a 80 µm long bridge. There are a number of possible approaches. first the implicit formula. The silicon nitride was deposited on a 2 µm thick polyimide layer on silicon substrates. (1) The analytical theory above could be redone adding two more sections to the beam modeling the moving supports. The uncertainty was estimated by performing repeated measurements. currently maintained by Landon Curt Noll. Deflections measured by the two Dektaks using a force of 24 mg agreed within Calc is an arbitrary precision calculator program. The factory representative could not give an accuracy for the setting. the beam will twist to some degree. SEM image of fixed–fixed silicon nitride beams with width of 19 µm and lengths from 120 µm to 200 µm. Figure 3. A program could be written to automate this process. which has a computer adjustable force setting between 0 and 50 mg. Figure 3 is an image from the Dektak 1 Figure 2. The ‘upwards’ pointing triangle below this is the probe’s shadow. 8]. The thin film beams tested here were made of silicon nitride grown using plasma enhanced chemical vapor deposition. This is similar to the method used to fabricate MEMS RF switches [7. This force setting was calibrated at the factory. The probe tip is the bottom of the dark triangle in the top part of the image. equation (4). This end effect must be dealt with in order to fit the theory to the data. and the polyimide was etched out from under the beams. Data reported in this paper will be for two silicon nitride samples. camera at the beginning of such a scan. but supporting silicon nitride at the ends of the beam is also undercut some distance. The lightest colored regions of the micrograph are where the polyimide has been etched from under the silicon nitride.6 µm thick film with a silicon rich composition. 2 Dektak is a registered trademark of Veeco Instruments Inc. Ts . Then the silicon nitride was etched into the beam shapes. (3) The method used here is to simply choose an effective length. 2 2 (12) 3. the value of x in equation (6) will always be Q. An example of some of the fabricated beams is shown in figure 2. for the beam. resulting in the measured deflection being too large. 688 . The other instrument was a Dektak 3ST. The calculations were done using a program called Calc1 . Two surface profilers were used. This was measured to be 24 ± 2 mg. Data analysis In order to calculate a deflection. Three methods of measuring and fitting the data to the model were used and are described in the next three subsections. A plot of a Dektak scan is shown superimposed on a micrograph of the beam in figure 4. (2) An approach similar to that used by Zhang et al [4] where the end supports were modeled with finite element numerical methods. The view provided by the instrument is not highly magnified and it is difficult to set the probe run precisely along the center of the beam. This means that the anchors at the ends of the beam are not rigidly clamped as was assumed in the preceding theory. This is due to the fabrication method and cannot be avoided. It should be noted that the Dektak was designed to give accurate height measurements and not use a particularly accurate force.1. Since the surface profiler measures deflection at the point where the force is applied.M W Denhoff + 2 C2 C2 2 (L − Q) + C2 (L − Q) − 2 2 sinh k(L − Q) 2 k + 2C2 D2 cosh k(L − Q) − 2C2 D2 − C2 D2 cosh2 (k(L − Q)) + C2 D2 D2k2 D2k + 2 sinh k(L − Q) cosh k(L − Q) − 2 (L − Q). One is a 1.isthe. It was not clear whether the uncertainty was due to the gauge or variability in the actual landing force of the probe. http://www. The nonlinear (Marquardt-Levenberg) fitting routines in the data analysis and plotting software available in our laboratory are not able to solve an implicit equation. using a simple wire spring force gauge. Scans along the length of a beam One method of data collection is to run the Dektak trace along the length of a beam. equation (6). It is licensed under the GNU Library or Lesser General Public License (LGPL). If the probe is not centered on the beam. can be found. The other is a 2 µm thick film with close to stoichiometric composition. The output from Calc was plotted with the data and the fit judged by eye or by calculating the root mean square deviation between the calculation and the data.

Tr =0. This value was subtracted from the force data before plotting in figure 8. As can be seen in the deflection curves in figures 4 and 5.0031 N E=10 GPa. but even if they were reliable they are not suitable for the determination of both E and Tr . This assumes that the moving part of the support has the same E and I as the beam. deflections of a single beam can be measured for a range of force values.0 × 1011 Pa and Tr = 4.8 µN (1. Thus.0071 N 0 50 0 Beam deflection (nm) -100 Beam deflection (nm) -50 -100 -150 -200 -250 -300 -350 -400 -450 100 120 140 160 80 µm beam E=1.Young’s modulus and residual stress 100 Data for 55 µm beam E=140 GPa. since the supporting silicon nitride should be stiffer than the patterned beam. other thin film samples that were tested did not fit well at the ends of the deflection. Fits to the actual data extrapolated to a force value of 12.9e11 Tr =2. The issue of determining Young’s modulus and the residual stress from a lengthwise scan will now be discussed. Using the Dektak 3ST. The undeflected beam height is found by measuring height of the beam supports. In other words. Good fits are achieved for E = 140 GPa and Tr = 0 N as well as for E = 80 GPa and Tr = 0. the effective length of the data was always the same as the distance between the tips of the polyimide.0031 N. Examples of the Dektak data are shown in figure 7.2. Data for three beams are plotted in figure 8. the length scan data are not reliable. The height measured is the height of the beam above the substrate with the force of the probe acting on the beam. A plot of scan data and fits to the model for a 55 µm long beam. It is surprising that the model fits the data near the ends of the deflection. it equals the sum of the polyimide thickness plus the silicon nitride thickness. As can be seen. The data for the 180 µm beam can be fit with E = 1. the beam will twist. An image of the Dektak probe about to scan across a 75 µm long beam. it will be stiffer than the beam. Tr pair cannot be determined. The scale of the x-axis is the same as that of the micrograph. This was interpreted as a zero offset in the force calibration of the Dektak. a unique E. and should have a larger value for I.000 N E=80 GPa. Tr =0. As mentioned above. A plot of scan data and a fit to the model printed on top of an optical micrograph of the actual 19 µm wide by 80 µm long beam. It is possible to align the probe to be within Figure 6. As the probe climbs up onto the beam. It seems that the Leff approximation is reasonable. A good fit of the model to the data is shown in figure 4 using Leff = 101 µm. The beam will level off as the probe passes the center and it will twist the other way as the probe runs off the beam. about 5 µm of the center of the beam.3 mg) for zero deflection. In order to determine the value of the beam deflection. In fact. Figure 6 is an image of the probe during a widthwise scan. Scans across the beam as a function of force Another method of collecting data is to scan the profiler across the center of a beam. Dektak probe position (µm) Figure 4. Now if E 689 .6 × 10−3 N with a root mean square fitting error of 7 nm. The error caused by this assumption will be small if the beam is much longer than the amount of undercut of the support. the height at the center of the scan is the wanted value. Tr =0. this height must be subtracted from the undeflected beam height. this length corresponds to the distance between the tips of the underlying polyimide that is supporting the beam. Regardless of the quality of the fit to the model. This distance will be used as Leff . Scan data and fits to the model for a 55 µm beam are plotted in figure 5. From these data alone. It is clear that since the support is wider than the beam. The good fit for this particular sample is likely due to it having a relatively small Young’s modulus and so the the stiffness is relatively less important than the tension. the bottom of the curve is relatively flat and a 5 µm offset would not yield too large an error.5e-3 L=101 µm 180 200 220 240 260 280 300 -200 -300 -400 -500 0 20 40 60 80 100 120 Dektak probe position (µm) Figure 5. This choice of parameters also fits the data for 105 and 80 µm long beams. 3.

Errors in the Dektak data will be discussed in the next section.2 × 1011 Pa and Tr is varied to give the best fit. while changing E mainly causes a shift along the x-axis of the ‘linear’ part of the curve. The original idea was to be able to measure Young’s modulus and the stress in different areas of the substrate to investigate the .6 × 10−3 /(1. can be converted to stress by dividing by the cross-sectional area of the beam.M W Denhoff 3500 3000 2500 2000 1500 1000 500 0 -500 100 L=105 µm F=10 mg L=105 µm F=25 mg L=300 µm F=15 mg 3500 3000 2500 2000 1500 1000 500 0 120 140 160 180 200 0 50 100 150 200 250 300 350 F=24 mg E=0. There is still some uncertainty in choosing the best values for E and Tr .15 mN. Dektak scans across beams of different lengths using different forces.24) mN. The results for scans across the beams measured with the Dektak 3ST and analyzed as a function of length are shown in figure 10. is set equal to 1.4e-3 Distance (µm) Figure 7. The curves are nonlinear for short effective lengths and tend toward straight lines for large effective lengths.4 nm). The effects of varying the two fitting parameters are such that it is easy to find unique values for each.3e-3 Length=105 µm E=1.4e-3 F=20 mg E=0.0e11 t=4. This requires good data with very little scatter. Deflection versus beam length data and fits for five forces. The accuracy is also limited by the 10% uncertainty in the force calibration.2e11 t=4. Deflection versus force data and fits for three beams with different lengths. Figure 9 plots the data against effective beam length.9e11 Tr =4.0e11 t=4. The sample was fabricated with four copies of the beam sets. This slight change in curvature is the basis for being able to find unique values for E and Tr .3 × 10−3 N with a fitting error of 12 nm. it runs through the bottoms of the small L data points. The best fits for this sample give E = 90 GPa ± 10% and Tr = 4.6e-3 Length=80 µm E=1. The uncertainties are the root mean square deviation of the five values. 3.3.4. The data for 10 and 15 mg force settings have two fitted curves which represent the range of possible fits.0e11 t=4. The longest beam has been pushed down to touch the substrate. but this is due to errors in the data. The second pair of parameters gives a worse fit and. Deflection (nm) 1400 1200 1000 800 600 400 200 0 0 50 100 150 200 250 Corrected force (µN) Figure 8. where it is difficult to judge whether a small change in E or Tr improves the fit.e.9e11 Tr =4.65e-3 F=10 mg E=0. that is. 3. with the larger slope.6e-3 E=1. giving worst case uncertainties of E = ±20 GPa and Tr = ±0.4e-3 F=15 mg E=1.9e11 Tr =4. E was adjusted to match the large L data so. This gives a residual stress of σr = 4. then Tr = 4. as can be seen in the graph. Scans across the beam as a function of beam length Data for the deflection of the center of the beams can be analyzed in another way. The difference in these two fits represents the uncertainty in the fitting process. Results for a second silicon nitride sample A 2 µm thick silicon nitride film with approximately stoichiometric composition was also measured.9e11 Tr =4.9 × 10−5 )(1.1e11 Tr =4.4e-3 F=5 mg E=0. as a function of beam length for a constant force. The (dashed) curve with E = 90 GPa and Tr = 4. The first pair clearly gives a better fit.9e11 Tr=4. Using these values for the fitting errors assumes that the Dektak measurement errors are random.08). they result in slightly greater curvature in the calculated line (dashed). would be about half of the differences between the above best and worst case fits (i.03 ± 0. The residual tension.5 mN (root mean square fitting error = 9. When fitting this curve. The model curve with E = 110 GPa and Tr = 4. 2000 1800 1600 Length=180 µm E=1. A reasonable fitting uncertainty. The effective length is the actual beam length plus 21 µm. The best values of E and Tr for these is an average of the two fits.6 × 10−3 N ± 2% where these are the uncertainties due to the fitting quality. 690 There are two model curves plotted for the data with a 15 mg force setting. The average over the five data sets is E = (184 ± 10) GPa and Tr = (4. Changing Tr mainly affects the slope of the ‘linear’ part of the curve. E = ±10 and Tr = ±0.65 mN is the best fit (root mean square fitting error = 5.0 nm) has a slightly higher slope due to the smaller value of Tr . which has units of force.6e-3 Deflection (nm) Height (nm) Effective length (µm) Figure 9.5e-3 E=0.6 × 10−6 ) = 151 MPa. Repeated measurements were done to test the source of errors using the Dektak 3 (with the fixed force setting).

. done on different days.9e-3 The results from figure 11 are E = (170 ± 30) GPa and Tr = (4. In the third method.3e-3 E=1. stoichiometric silicon nitride film. the tensile stress can be calculated to be σ = (113 ± 30) MPa. inhomogeneities or defects. data could be analyzed in terms of simple bending theory.4e11.1e11. The variations in the deflection measurements at each length are on the order of 100 nm.5e-3 F=10mg E=1. It also includes single sets of measurements on two other sets of beams (on the same substrate). then.3e-3 F=15mg E=2. Tr =3. Deflection (nm) 4. Plot of the results of repeated measurements on one sample. Tr =5.6e11. Tr =3. There are two experimental difficulties. The residual stress in PECVD SiN films can typically range from 400 MPa compressive to 140 MPa tensile and the Young’s modulus from 85 to 210 GPa [9]. Two other possible reasons could be variations in the force applied by the Dektak probe or some inconsistency in the actual movement of the beams.1e11.7e11. the repeatibility of E and Tr for SiN films is similar to the accuracy of the measurements reported in this paper. The goal is to have a relatively easy and inexpensive determination of both Young’s modulus and the residual stress. Tr =4.0e-3 E=2. the deflection is plotted as a function of the length of the beams.3) mN. When a solid structure of similar height was measured repeatedly with the Dektak. It would seem that the variations are due to the Dektak measurement itself. Typically. The analysis of this situation would be complex. A best fit of the model to the data and ‘minimum’ and ‘maximum’ possible fits are shown. The effective length is the actual beam length plus 15 µm. likely involving numerical modeling of a three dimensional problem. 1000 900 800 Beam Set A 25 Jun Beam Set A 30 May Beam Set A 7 Jun Beam Set B 7 Jun Beam Set D 24 Jun E=1. The errors are from the maximum and minimum fits. uniformity of the deposition. One possible method to obtain a separate value for E would be to measure the deflection of a cantilever. The determination using the Dektak 3ST looks more precise because results from five different force settings were averaged. Is the level of accuracy obtained in this measurement useful for PECVD films? In general. the residual stress will be zero and the bending motion will not involve any stretching of the beam. individual measurement fluctuations were similar for the two Dektaks. so this possibility was not tested. varying E or Tr has quite distinct effects on the theoretical curve. Tr =3.3e11. one needs better data than the Dektak measurements.Young’s modulus and residual stress 1800 1600 1400 F=24mg E=1.5e-3 E=2.1e11. The problem is that it is difficult to maintain constant film properties from run to run. The values for E and Tr determined using the two Dektaks agree within the 10% force calibration uncertainty. a much smaller force is needed to deflect a cantilever compared to a fixed–fixed beam. Tr =4. Knowing the cross-sectional area of the beams (3. In principle. The variations in the measurements of the beam deflections were not. We had no independent method to measure the force on the probe. This agrees within errors with the stress of a similar film that was determined by a wafer curvature measurement to be (130 ± 20) MPa. Examination using an SEM before and after Dektak scans did not reveal damage or any change in the beams. directly due to the height measurements. a 10 µm long cantilever will be bent down to touch the substrate. The effective length is the actual beam length plus 15 µm. Tr =4.9e11. Tr =5. Since the cantilever is fixed at only one end. With the films used here. on one set of beams (labeled set A). The advantage is that it is possible to tune the film properties for a specific application. The first two methods rely on fitting small changes in the shape of the plotted curve to obtain values for both E and Tr . The other problem is due to the undercut of the beam support during microfabrication. A statistical analysis of the fit would give smaller error values (roughly half of these values) and would be appropriate if one was using this measurement to compare two samples rather than finding an absolute measurement of E and Tr .1e11.6e11. However. Inconsistent motion of a beam might occur due to curling. Tr =4. Results for a 2 µm thick. However. In this case. Since there is no stretching.6 ± 1. Tr =3. This allows a fairly independent determination of E and Tr from the data. It seems that variations in the Dektak probe force is the most likely source of the deflection measurement random errors. Discussion Three methods of analyzing deflection data for fixed–fixed beams have been presented. The variation in the data shows no pattern with respect to the day of the measurement or the beam set.6e-3 F=20mg E=1.8 × 10−11 m2 ± 10%). Figure 11 shows the results for three sets of measurements.6e-3 E=2. The range of useful forces is smaller than the forces available with the Dektak. the fitting precision is better than the accuracy and this measurement would be useful to 691 Deflection (nm) 700 600 500 400 300 200 100 0 0 20 40 60 80 100 120 140 160 Effective length (µm) Figure 11. To do this. the properties of PECVD films are very sensitive to changes in the deposition conditions. variations of about 10 nm were obtained.8e-3 F=5mg E=1. The support will move during cantilever deflection and this ‘end effect’ will be important for beams that are not long compared to the undercut.2e-3 1200 1000 800 600 400 200 0 0 50 100 150 200 250 300 Effective length (µm) Figure 10.

IEEE Micro Electro Mechanical Systems Workshop (Napa Valley. Howe R T and Muller R S 1998 Surface micromachining for microelectromechanical systems Proc. Lett. Zhan M-H and Chen L-Q 2000 Microbridge testing of silicon nitride thin films deposited on silicon wafers Acta Mater. 48 2843–57 [5] Jensen B D. IEEE 86 1552–74 [2] Tai Y-C and Muller R S 1990 Measurement of Young’s modulus on microfabricated structures using a surface profiler Proc.M W Denhoff check for drifting properties of PECVD films. on Materials and Device Characterization in Micromachining II (Santa Clara. ANTEM 2000 (Winnipeg. 30 July–2 August 2000) pp 79–82 [8] Grant P D. Acknowledgments The author would like to thank Hue Tran and Milton Harry for their help in preparation of the SiN thin film samples. J. CA) pp 147–52 [3] Qin M and Poon V M C 2000 Young’s modulus measurement of nickel silicide film on crystal silicon by a surface profiler J. Micromech. Mansour R R and Denhoff M W 2002 A comparison between RF MEMS switches and semiconductor switches Can. and Jeff Fraser for the SEM analysis. Microeng. Eng. CA) SPIE 3875 61–72 [6] Landau L D and Lifshitz E M 1986 Theory of Elasticity 3rd edn (Oxford: Pergamon) [7] Denhoff M W. Su Y-J. References [1] Bustillo J M. Comput. Kronast W and M¨ uller B 1996 LPCVD against PECVD for micromechanical applications J. Qian C-F. On the other hand. the range of possible E and Tr values is very large and the accuracy of the measurements described here would be useful in developing a growth process to obtain specific film properties. 6 1–13 692 . Mater. Elect. Sci. Harry M A and Yu M 2000 Fabrication of a microwave MEMS switch Proc. Grant P D. Bitsie F and de Boer M 1999 Interferometric measurement for improved understanding of boundary effects in micromachined beams Proc. SPIE Conf. 19 2243–5 [4] Zhang T-Y. Kov´ acs A. 27 pp 33–9 [9] Stoffel A.