CMOS Torsional MicromirrorEddie L.C. Ng and Kenneth T.Z. Oo7. ConclusionReferencesCMOS Torsional Micromirror Eddie L.C. Ng and Kenneth T.Z. Oo Department of Electrical Engineering and Computer Sciences University of California, Berkeley Email: [email protected], [email protected] Abstract This paper presents the concept and design of a torsional micro-mirror system in a standard multi-metal layer CMOS process. The necessary torsional force is provided by CMOS comb drives that actuate in z-direction. Simulations yield that at least static ±40 rotation angle can be achieved with less than 50V for an arbitrary 3-metal-layer CMOS process with 0.5-µm layer thickness. 1. Introduction Various torsional micromirrors have been studied and developed in silicon micromachining techniques. A typical MEMS torsional mirror is driven electro-statically by the electrodes buried in the substrate directly underneath the mirror. Magnetically driven torsional micro-mirror has also been demonstrated [1]. However, direct integration of these MEMS structures with electronics is desirable for systems with arrayed micromirrors and microsensors on a single monolithic chip. 100%-CMOS-compatible micromirrors will lower manufacturing costs and enable monolithic fabrication on typical CMOS chips. On-chip electronics also removes the interconnect bottleneck since active devices can be placed close to the microstructures. A compelling example of this revolutionary CMOS-compatible micromirror technology would be the Digital Micromirror Device (DMD) invented in 1987 by Larry J. Hornbeck at Texas Instruments [2],[3]. DMD is a spatial light modulator in which an active micromirror is built on top of a CMOS SRAM cell. Electrostatic forces based on the data in the memory cell tilt the mirror either ±10 degrees to act as an ON and OFF switch for modulating the incident light. Although the DMD is fabricated by CMOS-like processes over a CMOS memory, the required fabrication steps are relatively complex and not readily available to general users [4]. Unlike the DMD in the TI’s CMOS-like processes, the torsional micromirror investigated in this paper can be fabricated using more conventional multi-metal layer CMOS processing followed by a sequence of simple maskless dry-etching steps. The CMOS-MEMS process developed at Carnegie Mellon University is a maskless, post-CMOS micromachining in which the etching masks are provided by the interconnect metal layers in the standard CMOS process [5]. Taking advantage of the multi conducting layers readily available in CMOS processes, we propose a purely CMOS torsional micromirror. The mirror is electrostatically driven by the z-axis comb-finger actuators. In this paper, the post-CMOS micro-machining is briefly described first, followed by the principle of z-axis comb-finger actuation. Design issues and tradeoffs are discussed and simulation results are analyzed. Several test structures are proposed to verify the expected results. 2. Fabrication and Process Flow The process flow is shown in Figure 1 to illustrate how our micromirror and comb drives can be fabricated using conventional CMOS processes. A thick top metal layer is used as the mask during the post-CMOS anisotropic CHF3/O2 Reactive IonEtch (RIE), which has a high selectivity of dielectric to metal, to define the microstructures [5]. This step etches oxide under the contact holes all the way down to the substrate, allowing the use of a dry SF6/O2 [5] or XeF2 plasma etch [6] for the next step that isotropically undercuts the SCS substrate and releases MEMS structures. This technique clearly has advantages. First, post-CMOS process is maskless. Second, the fact that dry etches are employed avoids the problem of potential adhesion of MEMS structures to the substrate. And lastly, the embedded multiple conducting layers in the structure enable electrostatic actuation in all x, y, and z axes [7],[8]. A simple Cadence layout of our micro-mirror is shown below in Figure 2. Figure 2: Simple Cadence layout of the micromirror in a 3-metal layer process. Note the use of vias to connect metal layers within the mirror while stator electrical layers are isolated electrically. 3. Design Theories The basic concept of how z-axis actuation can be realized is depicted in Figure 3(a) and 3(b). One can start with the biasing configuration of the conducting layers as shown in Figure 3(a). All metal layers on the moving fingers are electrically connected to ground while all metal layers on the stators are separately biased. The CMOS comb drive in this configuration is equivalent to a conventional lateral-axis polysilicon comb drive. When the bias voltage is removed from the metal-1 layer on the stators on one end of the mirror, a change in sidewall capacitance between the fingers occurs. This change in capacitance leads to an upward actuation of movable mirror fingers in z-direction and consequently the torsional rotation of the mirror itself. We also switch the bias and ground the metal-3 layer on the stators on the other end of the mirror to produce downward motion at the same time to double the torque. Figure 3(a): Close 3-D view of the comb drive actuators located on each side of the mirror. Figure 3(b): Electrode voltage scheme which provides vertical lifting force to rotate the mirror. However, evaluating the change in sidewall capacitance with respect to z-displacement of the fingers is a nontrivial task. We hence used FastCap [9], a finite boundary element solver developed at M.I.T. to calculate 3-dimensional self and mutual capacitances of conductor surfaces. Model used for FastCap simulation is depicted in Figure 4 and results are shown and discussed in the later section. KnowingdzdC, the vertical electrostatic force that is acting on each end of the mirror is given by 221VdzdCNFzgez= (1) where Ng is the number of comb-finger gaps and V the applied voltage. Assuming forces with the same magnitudes but in opposite directions on each end of the mirror (which will not be the case for the metal layers with different thickness), the resulting moment of the mirror about its support beam is aFaFMezeze==22 (2) where a is the length of the mirror side that is normal to the axis of rotation. The elastic recovery torque Mr of the torsion beams (with width w, length l, and thickness t) is θβθθ==lGwtkMr323 (3)where
View Full Document
Unlocking...