We present the novel 3D wafer-level vacuum packaging technology for MEMS, which only needs two masks and is realized by Au-Si eutectic bonding with a low-resistivity Si Cap wafer. Wherein, Au-Si bonding pads and rings are formed to accomplish vertical interconnects and a vacuum package respectively. Applied to resonators with two typical substrate wafers of Si and Glass, this 3D package structure has further gone through the systemic characterizations. Specifically, the high-reliable mechanic strengths of device packages are revealed from Au-Si bonding strengths measured as about 20 MPa in tensile tests. Based on analyzing parasitic capacitance networks in device packages, the feed-through effect in resonance tests can be effectively relieved by grounding on each Si layer in device package structures (i.e. the parasitic capacitance reduced from the level of pF to fF). By resonance tests, it is found that increasing the degassing duration (30 min $\to 60$ min $\to 100$ min) is beneficial for improving package vacuum degrees as well as consistencies (69% within 4–30 kPa $\to 88$ % within 0.6-1.2 kPa $\to 85$ % within 0.6-0.8 kPa), and hermetic package yields (73% $\to 86$ % $\to 100$ %). Besides, device packages are verified to have good long-term stability by monitoring quality factor variations in 6 months. Overall, this proposed 3D packaging technology has extensive application prospects in various MEMS devices due to its inherently-simple process, low capacitive feed-through effect, and fine package performances. Also, test results provide instructions on the realization of low-parasitic-capacitance signal interfaces and the optimization of wafer-level vacuum package processes.

Driven by ever-growing demands on 3D integration, various state-of-the-art electronics packaging techniques have been developed. This study presents a novel and cost-efficient 3D wafer level packaging technology based on co-planar Au–Si bonding structures, whose two remarkable features are (1) eliminating the height difference derived from multi-layer metal interconnection lines crossing bonding rings by building co-planar bonding structures and (2) accomplishing vertical interconnections of low-resistivity Si column structures by forming Au–Si bonding ohmic contacts. In this paper, the packaging structure is interpreted by designs, fabrications and tests, in which the efficiency verification on co-planar bonding structures and vertical interconnections is concretely addressed. The performances on co-planar bonding structures have been revealed by 3D profiles of bonding surfaces, cross-sectional SEM images (the gap-free bonding layer) and tensile tests for Au–Si bonding strength. Besides, tests on leak rates of packaged chips indicate good packaging hermeticity. In terms of vertical interconnection properties, an in situ extracting method on the specific contact resistance ( ρ _c ) is also introduced to quantitatively appraise the quality of Au–Si ohmic contacts. Therefrom, the measured resistance of a vertical interconnection (~1 Ω) could be qualified for most devices’ interconnection requirements. And further, the extracted ρ _c values of 1.83–3.48 × 10 ⁻⁸ Ω · m ² also imply forming of good ohmic contacts in Au–Si bonding. More importantly, being analogously conducted by any available eutectic bonding techniques, this 3D packaging method has inherently extensive application prospects.

We present a MEMS magnetometer based on the Faraday electromagnetic induction principle that works in the torsional resonance mode to sense in-plane magnetic fields. As for the device design and fabrication, a self-alignment technique is introduced to realize vertically-interlaced combs used for electrostatic driving for a torsional frame structure. In order to directly acquire relatively pure resonance characteristics of devices, an electromagnetic detection means with differential outputs is proposed to suppress the parasitic feed-through effect existing in the normal electrostatic detection. Based on magnetic response tests, the proposed device has the measured sensitivity of 491 mV/mT with the non-linearity of 0.40% and the resolution of 6 μT under atmospheric pressure. Featured with the quasi-zero power consumption (75 nW), the measurement bandwidth without mechanical bandwidth limits, and the wide dynamic range (98 dB), this in-plane-sense device constituting as an essential for a 3-axis MEMS magnetometer has wide application prospects in automotive and portable electronics

KOH etching and Au/Si eutectic bonding are cost-efficient technologies for 3D device fabrication. Aimed at investigating the process compatibility of KOH etching and Au/Si bonding, KOH etching tests have been carried out for Au/bulk Si and Au/amorphous Si (a-Si) bonding wafers in this paper. For the Au/bulk Si bonding wafer, a serious underetch phenomenon occurring on the damage layer in KOH etching definitely results in packaging failure. In the microstructure analysis, it is found that the formation of the damage layer between the bonded layer and bulk Si is attributed to the destruction of crystal Si lattices in Au/bulk Si eutectic reaction. Considering the occurrence of underetch for Au/Si bonding must meet two requirements: the superfluous Si and the defective layer near the bonded layer, the Au/a-Si bonding by regulating the a-Si/Au thickness ratio is presented in this study. Only when the a-Si/Au thickness ratio is relatively low are there not underetch phenomena, of which the reason is the full reaction of the a-Si layer avoiding the formation of the damage layer for easy underetch. Obviously, the Au/a-Si bonding via choosing a moderate a-Si/Au thickness ratio (⩽1.5:1 is suggested) could be reliably compatible with KOH etching, which provides an available and low-cost approach for 3D device fabrication. More importantly, the theory of the damage layer proposed in this study can be naturally applied to relevant analyses on the eutectic reaction of other metals and single crystal materials.

Restricted by the shapes of utilized resonance structures, conventional Lorentz-force MEMS magnetometers are rarely designed with the multi-turn coil layout for the performance enhancement of Z-axis magnetic sensing by increasing the effective coil length. We put forward a novel Lorentz-force MEMS magnetometer with a low-frequency annular resonance beam for Z-axis sensing. Wherein, four resonance elements are interconnected and coupled into the annular resonance beam with the contraction-expansion (CE) mode (~37.4 kHz), which contributes to generating together Lorentz forces on four sides of the MEMS magnetometer structure. Meanwhile, a three-turn Lorentz force coil with only two coil pads, which is utilized for injecting the alternating current, is located on and electrically insulated with the annular Si resonance beam. Under the Z-axis magnetic field, capacitance comb structures around the annular beam are employed to detect the resonance displacement induced by the Lorentz force. In this paper, the proposed MEMS magnetometer is illustrated with device designs, theory analyses, device fabrications, interface circuit realizations, and performance characterizations. Specifically, the designed MEMS magnetometer is fabricated using bulk Si micro-machining processes. In addition, a board-level interface circuit is realized by amplification, phase shift, demodulation, and filtering. By experimental tests, the fabricated MEMS magnetometer has a sensitivity of 214.2 mV/mT with a non-linearity of 0.42% and a resolution of ~3 μT estimated from noise measurements. Also, it exhibits fine performances on the cross-axis rejection ratios (37 dB), hysteresis (0.34%), and repeatability (~0.9%).

As the basis of various MEMS sensors and actuators, MEMS resonators are widely beset with the feed-through effect from parasitic capacitance networks, which makes it obstructive in directly acquiring the inherent resonance frequency and quality factor of a MEMS resonator from measured resonance responses. From the fresh perspective of resonance data modeling, we put forward machine-learning algorithm models to expediently extract these two critical indicators from resonance responses under varying-degree feed-through effects. Firstly, the resonance datasets, which contain numerous feature data from amplitude-frequency responses and target data of the two critical indicators, are constructed with the equivalent circuit model. By introducing the dominance analysis based on the Random Forest (RF) algorithm, the heavy-weight parameters are selected from feature data and further serve as prediction evidence for cost-efficiently extracting the inherent resonance indicators. Then based on the ~0.3-million-group datasets, two machine-learning algorithm models are trained with back-propagation (BP) neural networks for the two resonance indicators. Also, by optimizing hyper-parameters in training, the test results from testing datasets manifest fine accuracy in extracting the resonance frequency (0.05%) and quality factor (0.71%) for the trained algorithm models. Lastly, the experimental verification of trained algorithm models is implemented on a realistic MEMS resonator under four cases of feed-through effects, whose results indicate its low prediction errors of resonance frequency (<0.1%) and quality factor (<0.5%). In brief, the proposed algorithm models provide a novel strategy to accurately obtain inherent MEMS resonator indicators without any dependence on device structures, circuits and operation conditions.