Silicon and WLP enable commercial-grade MEMS resonators
06/01/2007
In today’s consumer electronics world, where integrated circuits provide the majority of system functionality, vibrating mechanical devices in the form of quartz crystals are still used as clock sources in most applications. To illustrate the stringent stability requirements for such systems, consider a device vibrating at 20MHz with a peak-to-peak displacement of 10nm. In a year’s time, a given point on the device travels a distance equivalent to a trip from New York to Tokyo. When you consider that typical frequency stability requirements are ≤50ppm, the allowable error over that distance is only a few hundred meters.
With the progress of microelectromechanical systems (MEMS), vibrating mechanical devices that meet these stringent requirements can now be manufactured from silicon using conventional surface micromachining technologies. When incorporated into a wafer-level packaging scheme, such MEMS resonators can provide a small, highly reliable, low-cost alternative to quartz that can also be integrated with an ASIC in the same low-cost package. Such systems have been demonstrated in consumer electronics applications.
Resonator process
A variety of different designs can be used to create MEMS resonators [1]. This article will focus on a design that uses a capacitively transduced, out-of-plane flexural mode (see “Resonator physics” on p. 64). This design has the advantages of low-voltage operation and a straightforward fabrication process. This type of resonator can be built using conventional silicon surface micromachining technology with two polysilicon layers. In surface micromachining, the mechanical structures are built from layers deposited on top of the substrate, in contrast to bulk micromachining in which structures are etched into the substrate, often tens or hundreds of microns deep.
An advantage of the surface micromachining process is that it uses process technology similar to mainstream silicon processing, allowing it to take advantage of established high-yield, low-cost processes and to be fairly portable. However, there are some important differences. For example, in addition to having good electrical properties, the selected materials must also have good mechanical properties. For example, the polysilicon films must have a consistent modulus of elasticity, low intrinsic stress, and good mechanical stability under aging in order to meet stringent frequency stability requirements.
The resonator process starts with a thermal oxide film to provide electrical isolation from the substrate. This is capped with a low-stress silicon nitride layer to protect it from the hydrofluoric acid used in the release process. A film of doped polysilicon is then deposited and patterned to form the lower electrode and interconnection layer (Fig. 1a). A thin layer of sacrificial oxide is then deposited to create the gap between the upper and lower electrodes. The thickness of this gap is dictated by resonator performance requirements. A thinner gap allows for greater coupling between the electrodes, and therefore a larger output signal.
Two sets of vias are fabricated next. One set of vias provides an electrical connection to the substrate, and the other provides the anchor points between the upper and lower electrodes (Fig. 1b). Another layer of doped polysilicon is then deposited (Fig. 1c) and patterned to form the upper, free electrode (Fig. 1d). Another sacrificial oxide layer is deposited to serve as a buffer between the upper polysilicon and the subsequent metal layer. Vias are etched through this layer for bond pads, which are then patterned and etched (Fig. 1e). The final step in the process is to remove the sacrificial oxide layers, thereby freeing the upper electrode (Fig. 1f). This release step is done using a vapor-phase hydrofluoric acid process (VHF) [2].
VHF has several advantages over the conventional wet release commonly used. First, it avoids the potential problematic attack of liquid hydrofluoric acid on metal bond pads. Second, it provides a dry-in, dry-out process, which avoids any potential issues with capillary forces and stiction, obviating the need for a post-release super-critical carbon dioxide drying step, which is often required after wet release. The result is a high-throughput, manufactuarable process that lends itself to release at the wafer-level.
Wafer-level packaging
After release, the resonator is vulnerable to a variety of environmental factors and must be packaged for protection. In addition to the typical microelectronic packaging requirements, MEMS resonators have another, more stringent requirement. Since the mass of a typical MEMS resonator is <10-10 kg, its performance can be strongly affected by the surrounding air molecules [3]. To mitigate these air-damping effects, vacuum packaging is required. It is highly desirable to do this at the wafer level in order to improve throughput and reduce cost.
In wafer-level packaging, a separate cap wafer is fabricated with an array of waffle-like cavities that will ultimately form pockets around the resonators. The two wafers are brought together under high vacuum using a variety of technologies and a combination of heat and pressure to form a seal [4, 5]. In addition to providing a sealed vacuum environment for the resonator, the chosen approach must be low cost, highly reliable, and enable the creation of small-sized packages. The sealing method must also be compatible with the backend processes, including thinning, dicing, and handling by automated assembly tools.
One technology that has been well-established to meet all these requirements is glass frit sealing, a technique that uses low-temperature glass mixed with binders and solvents into a paste that can be screen-printed using conventional methods. After screen printing, the frit is fired to solidify it and remove the solvents, leaving a film that can be bonded to form a vacuum seal with an ultimate pressure in the millitorr range.
A cap wafer with a glass frit sealing ring positioned over the mating MEMS wafer is illustrated in Fig. 2a. Figure 2b shows the same wafer after bonding and partial dicing of the cap wafer to expose the bond pads. An infrared transmission image of a bonded device is shown in Fig. 2c. Wafer-level packages fabricated in this way have been demonstrated to meet applicable bond strength and shear force requirements [6], which are critical to maintaining high yield during handling by the various automated tools used during assembly.
The vacuum level in which the resonator operates affects not only its performance, but also its long-term reliability. A time-honored approach to maintain the vacuum level is to use a getter. In the current application, a nonevaporable, high porosity metallic getter is deposited inside the cavities on the cap wafer after the printing and firing of the frit (Fig. 2).
The getter material consists of an alloy that is formulated to maximize the gas sorption performance while maintaining an activation curve that is process-compatible. The activation of the getter is a critical step because it removes the layer of surface oxides, nitrides, and carbides by diffusion into the bulk of the getter material, and provides a clean metallic surface ready to react with any impinging gas molecules.
Getter activation is performed during the sealing process using a suitable combination of temperature and time. After sealing, the presence of the getter ensures the long-term stability of the vacuum package through chemical sorption of active gases such as H2O, O2, CO, CO2, N2,and H2 [7].
Wafer-level packages fabricated in this way have been demonstrated to meet reliability standards, including accelerated moisture resistance testing using an autoclave (per JEDEC standard JESD22-A102), and high-temperature storage at 150ºC (per JEDEC standard JESD22-A103C) [6]. Because they enable the fabrication of many thousands of devices in parallel, wafer-level packages are a critical component in the manufacture of small, highly reliable, low-cost MEMS resonators.
 Figure 3. SEM image of a MEMS resonator in a vacuum-sealed wafer-level package integrated with an ASIC into a conventional ceramic package. |
Because of its small size, the wafer-level packaged MEMS resonator can readily be integrated with an ASIC into a single package. Figure 3 shows a vacuum-sealed resonator stacked on an ASIC and wire-bonded in a ceramic package. An inexpensive, conventional plastic QFN package can also be used, as the vacuum sealing required for the resonator is provided by the wafer-level package. A Discera oscillator manufactured in this way has been demonstrated in a high-performance camcorder that exhibited the same image quality and performance as with a stock quartz crystal oscillator.
Conclusion
MEMS resonators have been fabricated that meet consumer electronics performance requirements. These resonators leverage existing high-yield fabrication processes. When incorporated with a wafer-level packaging scheme, the result is a device that is small (<1mm2), reliable, low cost, and capable of being manufactured in high volume to meet the needs of the consumer electronics clock source market.
Acknowledgments
The author would like to thank the members of the Discera technical team and our manufacturing partners for all their efforts in making commercial grade MEMS resonators a reality. Special thanks to Minfan Pai and Dinh Vu for providing the images of fabricated devices.
References
- W.-T. Hsu, “Vibrating RF MEMS for Clock and Frequency Reference Applications,” Invited Paper, International Microwave Symposium, San Francisco, CA, June 2006.
- K. Torek, J. Ruzyllo, R. Grant, R. Novak, “Reduced Pressure Etching of Thermal Oxides in Anhydrous HF/Alcoholic Gas Mixtures,” Journal of the Electrochemical Society, Vol. 142, pp. 1322-1326, April 1995.
- Xia Zhang, W.C. Tang, “Viscous Air Damping in Laterally Driven Microresonators,” IEEE Workshop on Micro Electro Mechanical Systems, Oiso, Japan, pp. 199-204, Jan. 1994.
- Daniel N. Pascual, “Fabrication and Assembly of 3D MEMS Devices,” Solid State Technology, July 2005.
- George A. Riley, “Wafer-level Hermetic Cavity Packaging,” Advanced Packaging, pp. 21-24, May 2004.
- W.-T. Hsu, “Reliability of Silicon Resonator Oscillators,” IEEE Frequency Control Symposium, Miami, FL, pp. 389-392, June 2006.
- Richard C. Kullberg, “Processes and Materials for Creating and Maintaining Reliable Vacuum and Other Controlled Atmospheres in Hermetically Sealed MEMS Packages,” MEMS Reliability for Critical and Space Applications, Proc. SPIE, Vol. 3880, pp. 75-82, Aug. 1999.
Barry D. Wissman received his AB in physics from Washington U. and his PhD in physics from the U. of Michigan. He is principal engineer at Discera, 655 Phoenix Drive, Ann Arbor, MI, United States; ph 734/528-6364, e-mail [email protected].
Resonator physics
Shown at right is a conceptual drawing of a basic capacitively transduced resonator. It consists of a fixed lower electrode and a free upper electrode separated by an initial gap d0. The free electrode has mass mr and is anchored via a suspension with stiffness kr. A dc bias voltage VP is applied to the upper electrode, which serves to charge it.
The application of an ac excitation signal vi to the lower electrode creates a time-varying attractive force. This causes the free electrode to move in a direction perpendicular to the fixed electrode. This movement is countered by a restoring force provided by the suspension. The interplay between these forces leads to vibrations with a natural frequency dictated by the free electrode’s stiffness-to-mass ratio. By adjusting these two parameters, the frequency can be tailored to match the desired application.
 Drawing of a capacitively transduced resonator. The lower electrode is fixed, while the upper electrode is free to move. |
As the resonator vibrates, the gap between the electrodes changes, and, as a result, the system behaves as a time-varying capacitor, which generates an output current proportional to the bias voltage multiplied by the time rate of change of the capacitance.
When the frequency of the ac excitation signal matches the natural frequency of the resonator, the free electrode is driven into resonance. A system driven in this manner responds much more strongly to frequencies close to its natural frequency. The measure of the sharpness of this response is the Q factor, defined as the ratio of energy stored to energy dissipated per cycle. Equivalently, Q can be expressed as the ratio of the center frequency to the 3dB bandwidth. The lower the damping constant (cr), the higher the Q. In operation, a resonator with high Q will be highly sensitive to the driving frequency.