Micro-mirror Arrays for Maskless Lithography B. J. Warlick and J. Garrett ABSTRACT This paper presents a design approach for fabricating an array of MEMS mirrors suitable for maskless lithography applications in a commercial CMOS process. A small footprint mirror design is presented which uses a pair of support beams for signal routing and diagonal gap-closing actuators for mirror deflection. The actuators are shown to be capable of delivering up to 10nN of force per micron of actuator for a supply voltage of 30V, and 0.1nN per micron for a thick gate oxide 0.25µm CMOS supply voltage of 3.3V. INTRODUCTION Maskless lithography has been proposed as a means to greatly reduce non-recurring engineering (NRE) costs in the fabrication of future process generations [1]. Maskless lithography entails the use of an integrated CMOS/MEMS microchip with an array of actuated mirrors and circuitry to transfer pattern data to a target wafer. Like a standard extreme ultraviolet (EUV) mask, the positive pattern is reflected onto the wafer (in this case, by mirrors in the “on” position). The negative pattern is reflected away by mirrors in the “off” position. To completely expose the wafer, the mask pattern will be electronically scrolled across the mirror array as the wafer is mechanically scanned under the focused pattern. Each pixel on the wafer is exposed by multiple mirrors, reducing the technique’s sensitivity to MEMS device yield. Current maskless lithography research at U.C. Berkeley is targeted for a process with a 50 nanometer minimum feature size, approximately 3 process generations from the current leading 130nm process [2]. We developed a micro-mirror that meets the maskless constraints for a 0.25µm process. A micron-scale, flexure based MEMS mirror has been proposed in [1]. The limitation of the flexure design is “snap down”: deflection is nonlinear and at about 1/3 of the gap, electrostatic forces will snap the mirror the remaining distance into contact with the base. Contact may result in wear, stiction, and charge accumulation. The flexure mirror avoids contact by just tilting enough to deflect the EUV beam away from the projection optics. In developing a physical model they assume that bending of the mirror is negligible—all bending occurs at the 100nm hinge/support. DESIGN The fabrication process used for our design was originally developed at Carnegie Mellon by Gary Fedder [3]. It is a CMOS compatible process with two post-processing steps. The first is an RIE etch of silicon dioxide down to either the first metal layer defined or the substrate. The second step is a wet etch of silicon dioxide. The wet etch releases metal structures that have been outlined by the RIE. An additional step is necessary beyond the normal Fedder process in which an EUV reflective layer is deposited on the mirror surfaces. This layer is described in [1], and we assume a process can be developed to deposit EUV reflective layers on our mirrors. The Fedder process possesses several qualities which make it highly attractive for maskless lithography: small feature size, the capability for integrated electronics, and high yield. Small feature size is inherent because the process scales with CMOS technology. The semiconductor industry invests heavily in new process technology to scale CMOS. The Fedder process leverages that scaling. Maskless lithography requires integrated circuitry for mirror control. The process is also high yield, when Figure 1. Micro-mirror layoutcompared to surface-micromachining post-processing steps. Several hundred million mirrors are required for rapid maskless lithography. Yield will not be one hundred percent for such a large array, but as many mirrors as possible must be functional. However, because the Fedder process has a small number of post-processing steps, the yield is high, which reduces costs. One significant drawback to design of micro-mirror arrays in the Fedder process is the inability to stack released MEMS structures above CMOS circuitry, something made possible in surface micromachined approaches. Methods for out-of-plane actuation compatible with the Fedder process have been demonstrated previously [4, 5]. The Fedder process is capable of producing multiple capacitors between comb fingers, as shown in figure 2. The three metal layers in the rotor are electrically connected. Metal-one and metal-three in the stator are separately connected. The configuration forms two sidewall capacitors. When a voltage is applied between the top or bottom capacitors, the rotor will shift vertically to minimize capacitance. The maximum displacement is a function of the vertical position that minimizes the activated sidewall capacitance. The out-of-plane actuation is not subject to snap down like the design in [1]. In [5], a torsion mirror is actuated by two out-of-plane comb drives. The stator comb fingers are defined by metal-three, metal-four, and metal-five. The mirror fingers are defined by metal-three. When a voltage is applied, the asymmetric comb drive creates an upward force on the mirror, causing it to rotate. Our basic design consists of a 5x5µm multi-layer square mirror (metal-one through metal-three, including underlying oxide layers), a pair of single-layer support beams (metal-one and underlying oxide), and a multi-layer stator (metal-one through metal-three, including underlying oxides) adjacent to one side of the mirror (figure 2). Mirror deflection is achieved by applying a voltage across pairs of diagonal gap-closing actuators (figure 3). The maskless lithography application requires a high mirror fill factor, meaning that the flexure must be small relative the mirror surface area, and a relatively square “footprint” for the combined mirror and support structure must be maintained. The support beams for our mirror are in an “antennae” configuration, anchored at the top of the mirror and turning backwards (Figure 1), in contrast to the single support design proposed in [1]. The use of antennae supports significantly increases the support beam length, while maintaining high fill factor for the mirror array. The spring constant of a support flexure is inversely proportional to the cube of the flexure’s length, an effect which more than compensates for the additive effect of introducing two beams in parallel – by decreasing the support spring constants, greater out-of-plane actuation can be achieved for less force. The
View Full Document
Unlocking...