Nanometer-Scale Flatness and Reliability Investigation of Stress-Compensated Symmetrically-Metallized Monocrystalline-Silicon Multi-Layer Membranes Mikael Sterner∗ , Göran Stemme∗ and Joachim Oberhammer∗ ∗ Microsystem
I. I NTRODUCTION Many MEMS applications, especially optical, require coated, typically metallized, membranes of high flatness maintained over the whole device lifetime. Purely metallic membranes, for instance aluminum mirrors, exhibit poor flatness and are susceptible to plastic deformation, especially at slightly elevated temperatures [1]. Monocrystalline silicon membranes, as employed by our research group for micromirrors [2] and for RF MEMS metamaterial surfaces [3], offer almost perfectly elastic mechanical properties and high flatness, but metallization layers, as required for reflective-surface applications, induce a stress gradient resulting in membranes bending and buckling. [4] Additional stress-compensating layers [5], [6], typically deposited on the same side as the reflective coating, or finely tuned deposition methods [7] are difficult to manage because they require accurate control of tolerances in the process parameters, and can only be optimized for a small operation temperature range. This paper presents a very robust stress-compensation concept for metallized monocrystalline silicon membranes, featuring symmetrical deposition of the metal on both sides of the membrane to balance the stresses induced by the metallization. The resulting multi-layer membrane retains the near-ideal mechanical properties of the monocrystalline silicon
z
direction of deposition
1
σ PROBLEM Si
Metal
Stress-compensation
Abstract—This paper demonstrates a very robust and fabrication-parameter insensitive concept of full stress compensation in metallized monocrystalline silicon membranes, by symmetrical metal deposition on both sides of a transfer-bonded silicon membrane, resulting in previously unmatched nearperfectly flat and high-reliability metal-coated membranes. Application examples are high-performance optical mirror devices and quasi-optical tuneable microwave surfaces. The influence of the thickness ratio between the metal films on the two membrane sides are investigated, demonstrating a controllable curvature range from −0.3 mm−1 to 0.1 mm−1 by varying the top to bottom metal thickness ratio from 0.38 to 3.5, using metal thicknesses from 200 nm to 800 nm, and achieving near-zero curvature down to 0.004 mm−1 . Extensive reliability tests, up to 100 million cycles, showed no detectable change in curvature, no plastic deformation and good repeatability in analog-mode deflection (within 2.5 %), proving the robustness of this concept of metallized monocrystalline membranes.
Conventional
Technology Lab, KTH – Royal Institute of Technology SE-100 44 Stockholm, Sweden, Email:
[email protected] 2
z σ SOLUTION
1
Fig. 1. Concept of stress-compensation by symmetrical deposition of metal (right), as compared to conventional metallized membrane with resulting stress gradient (left).
core, proven by life-cycle measurements over 100 million actuation cycles showing virtually no degradation or plastic deformation. In contrast to conventional stress compensation methods, where the coating layers must be thin compared to the structural layer, the presented concept is shown up to 800 nm of gold on a 1000 nm core. II. S TRESS - COMPENSATED SYMMETRICAL METALLIZATION OF MONOCRYSTALLINE - SILICON MEMBRANES The stress-compensation concept of this paper (Fig. 1) is based on depositing the metal layers before and after the transfer of the membrane to a target substrate, achieving metal layers that are deposited under exactly the same conditions, and from antipodal directions (i.e. on both sides of the silicon core the metal is deposited towards the surface). The fabrication process is outlined in Fig. 2. The thin device layer of a silicon-on-insulator (SOI) wafer is transferred to the final substrate by adhesive transfer-bonding at 200 ◦C with MR-I 9150XP sacrificial bond polymer allowing for subsequent release of the membrane in oxygen plasma. [8] The transferred monocrystalline-silicon device layer becomes the core of the multi-layer-metallized membrane, which is then patterned and supported by electroplated posts before the final release etch. The direction of the symmetrical metallization of the membrane is indicated by the arrows in Figs. 1 and 2. The first
Si
Au
SiO2
Sacrificial polymer
1
( c ) A d h e si ve w afe r b o n d i n
g
RIE (a1) SOI wafer
(b1) First metallization
1 (a2) Substrate
(b2) Pattern electrodes
(d) Strip handle and BOX
(e) Second metallization
(f) Pattern the membrane and electroplate posts
(g) Release the membrane by O2 plasma etching
2 1
Fig. 2.
Process flow outlining the fabrication steps of stress-compensated silicon membranes.
metal layer is deposited on the device layer directly before the transfer-bonding. The second metal layer is deposited on the other side of the device layer directly after it has been exposed by etching away the handle and the BOX layer of the SOI wafer. Since the metal layers are deposited under the same conditions, the stresses induced by the metallization will be balanced out by symmetry. This also makes the process very insensitive to machine parameter variations in the metallization step, such as stress and deposition rate, as long as the parameters are stable between the two runs on the same membrane. The final bending is independent on absolute deposition time, since only the ratio between the metal layer thicknesses affects the result. Fig. 3 shows SEM pictures of fabricated test devices, in this embodiment tuneable RF MEMS quasi-optical metamaterial surfaces [3], featuring 1000 nm thick silicon membranes covered by 500 nm of gold on both sides. III. I NFLUENCE OF THE THICKNESS RATIO OF THE METAL LAYERS
The matching of the thicknesses of the top and bottom metal layers is important for controlling the final bending of the membranes. In theory, if the deposition conditions are the same for the two metal layers, the ideal flatness should result from having the same metal thickness on top and bottom of the membrane. However, in our observations, the best flatness is achieved by having slightly less metal on the bottom. This could be explained by the internal state of the thin monocrystalline silicon device layer being different during the two metal depositions, for example caused by pre-stress from the SOI wafer or from the transfer-bonding. The final bending can, however, still be controlled to near-zero by tuning the thickness ratio of the metal layers. To investigate the influence of asymmetric thickness variations on the flatness of the membrane, test membranes were
electrode
moving membrane
anchor
spring
Fig. 3. SEM pictures of fabricated devices (tuneable metamaterial surfaces).
fabricated where the metal thickness on the top and bottom side were independently varied from 200 nm to 800 nm. All
ttop(x)
ttop
Fig. 4. Independent parameter variation of top/bottom thickness within the same wafer
0
600 500
−0.1
400
−0.2
300
−0.3 200
400 600 bottom thickness (nm)
0
800
Fig. 5. Curvature measurements for parameter variation of top and bottom metal thickness (total of 35 data points, each derived from 16 devices of same thickness combination).
−40
IV. M ECHANICAL RELIABILITY OF THE MULTI - LAYER MEMBRANES
The excellent mechanical properties of monocrystallinesilicon membranes are retained by the multi-layer membranes, since the highly elastic and stiff silicon core dominates over the softer metal layers. Fig. 6 shows white-light interferometric measurements of a stress-compensated test membrane, with a
0 position along scan (µm)
100
Fig. 6. Surface profile of a stresscompensated membrane obtained by whitelight interferometry, showing very uniform curvature distribution.
0.2
ba 3.6
0.1
3.2
0
2.8
−0.1
2.4
−0.2
0.5 µm Au 1.0 µm Si 0.5 µm Au
−0.3 −0.4
test membranes could be fabricated on the same wafer by employing a moving shadow-mask to create two stair-case patterns orthogonal to each other on the top and bottom side, as shown in Fig. 4. Fig. 5 shows the curvature distribution obtained by the independent parameter variation. The curvature is extracted from white-light profilometer measurements of 16 test membranes for each combination of top and bottom thickness. The isoline of thickness combinations resulting in zero bending is indicated in the figure, and it is clearly seen that this ideal line is offset from the symmetric case. However, the data can be used to select a deposition ratio of top to bottom metal thickness that gives a low curvature, which will then be independent of the exact processing conditions as long as the metal thickness is proportional to the deposition time.
a
−20 −100
curvature (1/mm)
700
0.1 curvature (1/mm)
zero curvature
20
0
10 20 actuation voltage (V)
30
2 1.6
ba 1.2
deflection of membrane (µm)
tbot
tbot(y)
deflection (nm)
0 µm
tbot
top thickness (nm)
a
z height
ttop
6 µm
Fig. 7. Curvature of the multi-layer membranes, invariant over actuated membrane deflection.
low and fairly uniform flatness of 20 nm to 40 nm over the 200 µm membrane, corresponding to roughly 0.01 mm−1 . b The test membranes are electrostatically actuated with rougly 30 % tuning range from the initial distance ofc 3.6 µm. In Fig. 7 the curvature and deflection is plotted for a symmetrically metallized membrane over the full analog tuning range, showing almost invariant curvature over the whole range (−0.165 mm−1 to −0.154 mm−1 ). Furthermore, the reliability of the membranes has been investigated by studying both the membrane shape and absolute deflection during life-cycle measurements over 100 million cycles (Fig. 8), demonstrating the excellent long-term b stability with virtually no degradation neither in curvature nor in repeatability of the actuated deflection. The curvature c varies by only 0.005 mm−1 and the deflection is within 2.5 % variation in both states. The actuation cycles were performed at 10 kHz unipolar frequency, and stopped at power-of-ten
c
curvature (1/mm)
0.1 0
−0.1
−0.161 −0.162 −0.163 −0.164 −0.165
2.7
0
1
2
3
4
5
6
7
8
10 10 10 10 10 10 10 10 10
−0.2 −0.3
Unactuated (0 V) Actuated (30 V)
−0.4 0 1 2 3 4 5 6 7 8 10 10 10 10 10 10 10 10 10 number of actuation cycles (a)
deflection of membrane (µm)
0.2
2.6 2.5 2.4 2.3 2.2 2.1
Unactuated (0 V) Actuated (30 V)
2 0 1 2 3 4 5 6 7 8 10 10 10 10 10 10 10 10 10 number of actuation cycles (b)
Fig. 8. Life-cycle measurements over 100 million actuation cycles, showing virtually no degradation or plastic deformation: (a) invariance of the membrane curvature; (b) excellent repeatability of the absolute membrane deflection. In the unactuated curve, both the measurement point before and the one after the actuated measurement are shown.
intervals to make a white-light interferometric measurement of the actuated and unactuated states. No charging was observed, since there is no dielectric isolation layer in the allmetal actuator. After 100 million cycles the measurement was stopped without observing any failure. This good reliability was achieved even though the experiment was performed with an unpackaged device in a general cleanroom area with no additional atmospheric control over temperature and humidity. V. C ONCLUSION A robust stress-compensation concept for metallized monocrystalline silicon membranes, featuring symmetrical metal deposition to balance stresses induced by the metal layers, has been shown to result in multi-layer membranes with near-zero curvature down to 0.004 mm−1 with up to 800 nm thick metal layers on a 1000 nm monocrystallinesilicon core, and near-ideal mechanical properties, proven by life-cycle measurements over 100 million actuation cycles showing virtually no degradation or plastic deformation. ACKNOWLEDGMENT The research leading to these results has received funding from the European Community’s Seventh Framework Programme FP7/2007-2013 under grant agreement n◦ 224197. R EFERENCES [1] M. Douglass and D. Kozuch, “DMD reliability assessment for large-area displays,” in Society for Information Display International Symposium Digest of Technical Papers, vol. 26 (Applications Session A3), May 1995, pp. 49–52. [2] M. Lapisa, F. Zimmer, F. Niklaus, A. Gehner, and G. Stemme, “CMOSintegrable piston-type micro-mirror array for adaptive optics made of mono-crystalline silicon using 3-D integration,” in Proc. IEEE Int. Conf. Micro Electro Mechanical Systems (MEMS), Sorrento, Italy, Jan. 2009, pp. 896–899.
[3] M. Sterner, D. Chicherin, A. Räisänen, G. Stemme, and J. Oberhammer, “RF MEMS high-impedance tuneable metamaterials for millimeter-wave beam steering,” in Proc. IEEE Int. Conf. Micro Electro Mechanical Systems (MEMS), Sorrento, Italy, Jan. 2009, pp. 896–899. [4] Y. N. Picard, D. P. Adams, O. B. Spahn, S. M. Yalisove, D. J. Dagel, and J. Sobczak, “Low stress, high reflectivity thin films for mems mirrors,” in MRS Symposium, vol. 729, 2002, p. U3.11. [5] Y.-H. Min and Y.-K. Kim, “Modeling, design, fabrication and measurement of a single layer polysilicon micromirror with initial curvature compensation,” Sensors and Actuators A: Physical, vol. 78, pp. 8–17, 1999. [6] K. Cao, W. Liu, and J. Talghader, “Curvature compensation in micromirrors with high-reflectivity optical coatings,” Journal of Microelectromechanical Systems, vol. 10, pp. 409–417, 2001. [7] A. Tarraf, J. Daleiden, S. Irmer, D. Prasai, and H. Hillmer, “Stress investigation of PECVD dielectric layers for advanced optical MEMS,” Journal of Micromechanics and Microengineering, vol. 14, pp. 317–323, 2004. [8] F. Niklaus, A. Decharat, F. Forsberg, N. Roxhed, M. Lapisa, M. Populin, F. Zimmer, J. Lemm, and G. Stemme, “Wafer bonding with nano-imprint resists as sacrificial adhesive for fabrication of silicon-on-integratedcircuit (SOIC) wafers in 3D integration of MEMS and ICs,” Sensors and Actuators A: Physical, vol. 154, pp. 180–186, 2009.