A Microelectromechanical Switch for Bandpass Vibration Detection

Wireless sensor networks continue developing towards spatially distributed nodes and the integration of a larger number of sensors. Efforts to reduce the resulting high data rates and power consumption include sleep/wake-up modes and asynchronous communication protocols. Distributed sensing and local data pre-processing, where only relevant information is disseminated through the network is often seen as the way to proceed. On the level of the sensor node the goal is autonomous operation powered by batteries and energy harvesters. This development has implications for the sensors integrated in such networks. They need to be low cost, low power and easy to integrate with digital circuits. This consideration shifts the demand away from high performance, high data rate, sensing that is local and of short duration to low power, always-on sensor front-ends. Commercial capacitive microelectromechanical system (MEMS) accelerometers offering these features are readily available. However, they typically cover the frequency range below 1 kHz. Structural Health Monitoring could benefit greatly from MEMS sensors, but higher bandwidths are required. There, commercial, sensitive piezoelectric devices are bulky and high priced while MEMS accelerometers consume too much power at low sensitivities.

To address this demand, the feasibility of a MEMS for low kHz vibration sensing is explored in this thesis. The system should integrate the functionality of a bandpass filter, signal amplifier and threshold detector. This approach reduces the need for continuous sampling at high frequencies, thus enabling vibration detection at minimal power consumption. Sensor output levels large enough to wake up a micro-controller in sleep mode are targeted. To achieve this, the approach of a two-stage transduction principle is followed: First, the vibration is dynamically amplified in a defined frequency range. This is done mechanically using a chain of coupled mass-spring-resonators. Thereby, amplified displacements and frequency selectivity are achieved at zero power expense. For the second stage, viz threshold detection, an electromechanical, dynamic, pull-in switch is chosen. When the output of the mechanical amplifier exceeds a threshold, an electrical contact is closed. While consuming only leakage current in off-state, comparably high output currents are achieved in on-state.

In this work, several steps have been taken to translate this concept into reality: First, the transduction principle is developed. This results in a lumped element model for a generic broadband pull-in vibration trigger consisting of n-resonators. The model is then translated into a planar design for fabrication in Silicon on Insulator (SOI) based micromachining. Second, a mechanical amplifier based on six resonators coupled in series is fabricated and characterized using a differential capacitive readout. A bandpass response from 3 kHz-13 kHz is achieved at a minimum and mean mechanical amplification of 16.4 dB and 23.5 dB, respectively. Subsequently, two types of Silicon-to-Silicon pull-in vibration switches are fabricated using the state of the art, near-vacuum epitaxial sealing (epi-seal) process developed in the Stanford Microstructures & Sensors Lab. The first type of switch is based on a single oscillating mass and the second on a dynamic amplifier with a chain of three resonators. Both devices only require a static voltage of 5.3 V for operation and are tested using a time domain approach. They are used to detect more than 23'000 vibration events each. This is done at an on/off power ratio of >1000 (11 nW in off-state vs. 13.6 µW in on-state). Further, the dependence of detectors' output states (open and closed) depending on the vibration's amplitude and frequency is investigated. The corresponding switching map of the dynamic pull-in is simulated and measured for a parameter space in frequency and amplitude. Close to the resonances, tongues of instability (trigger on) are observed, similar to literature. The response of the one mass trigger at the primary resonance has a minimum detected threshold of 7.7 4.1 nm. A broadband response with multiple overlapping tongues is achieved using the three-resonators device with a bandwidth of 6-18.7 kHz at 5.3 V. The minimum and broadband detection threshold is 1.1 nm 0.2 nm and 17.5 2.2 nm, respectively. One major advantage of the operation in the strongly nonlinear regime is the lowering of the broadband threshold (factor 6 compared to the linear case in the three resonators scenario). However, this also introduces a more complex switching behavior. Within the coarse, tongue-shaped envelope, a fine pattern of alternating bands of stable and unstable parameters is resolved. Similar patterns have been predicted numerically in literature for the dynamic pull-in of MEMS resonators. Here, they are also experimentally demonstrated.

Finally, predictions are made for the performance of future device generations. As a next step, increasing the number of resonators in the mechanical amplifier and scaling down the capacitive gap can reduce the triggering thresholds at no additional power expense. We expect median broadband thresholds of 56 pm for a device that can be fabricated with existing processes. Medium term goals include an ultra-low power readout circuit, vacuum packaging of the up-scaled designs, and the optimization of the acoustic interface to the surface under test. In conclusion, the operation of a microelectromechanical broadband pull-in vibration trigger was successfully demonstrated, opening a path towards ultra-low power broadband vibration detection. The device only requires a static voltage at low leakage for operation and provides output levels high enough to wake up a microcontroller in sleep mode.