MEMS accelerometers have been widely used in various applications such as automotives, consumer electronics, industrial sensing, and health monitoring. The signal bandwidth in these applications varies significantly, as shown in Table 1. For example, the acceleration frequency range in automotive application is up to 400 Hz (Yazdi et al., 1998). In human activity monitoring, the acceleration signals are typically sampled at 25–40 Hz (Mathie et al., 2001; Bulling et al., 2014). For physiological signals, the normal respiration rate is 0.1–0.5 Hz (Morillo et al., 2010) while the cardiac signals has a bandwidth of 5–35 Hz (Kwon et al., 2011). Accelerometers are also used to measure heart sound between 20 and 500 Hz (Hu and Xu, 2012). Therefore it is desired to have a sensor whose characteristics such as the bandwidth and sensitivity can be reconfigured/tuned depending on the requirements from specific applications.

Tunable MEMS accelerometers have been demonstrated over the past years. Most tunable accelerometers use electrostatic force to tune the effective spring constants in the device. For example, SOI technology was employed to implement an in-plane tunable resonant accelerometer where back-to-back tuning electrodes were used to enhance the sensitivity via the electrostatic spring softening effect (Xiong et al., 2021). For a bias voltage of 67 V, the sensitivity was increased from 492 Hz/g to 2,277 Hz/g in static tests. In (Daeichin et al., 2019), a two-layer polysilicon surface micromachining process was employed to fabricate a torsional out-of-plane accelerometer. In contrast to the more commonly used electrostatic attractive force, electrostatic repulsive force due to a bias voltage of 40–60 V was used in this device to tune its resonance frequency and sensitivity without suffering from the pull-in effect. In (Fain et al., 2018), a standard MEMS fabrication process was used to fabricate an in-plane rotational accelerometer. The tuning voltage up to 4.5 V was applied to the seismic mass through a RC network to control the sensing bandwidth. The resulting electromechanical torque effectively changed the quality factor and inertia in the equation of the motion of the sensor.

The tuning electrode configuration in tunable accelerometers based on the spring softening effect can be roughly divided into two categories: comb fingers for in-plane movement/sensing (Park et al., 1999; Zotov et al., 2015; Ding et al., 2017) and parallel-plate electrodes for out-of-plane movement/sensing (Lee et al., 2000; Caspani et al., 2014). In (Guo et al., 2021), both comb fingers and parallel-plate capacitors were used to control the in-plane stiffness-tunable accelerometer. As will be discussed in the following sections, the tunability and tuning voltage of electrostatic tuning are closely related to the gaps of the tuning electrodes. The smaller the gaps, the lower the tuning voltage and the easier the accelerometers can be tuned. For comb fingers, the vertical gaps are typically defined by photolithography and etching processes. In contrast, the gaps in parallel-plate electrodes are the thickness of deposited thin films if the gaps are formed by removing the sacrificial layers. Even though these parameters depend on specific device design and fabrication processes, the parallel-plate electrodes adopted in this study would have potentially smaller gaps and better tuning characteristics than the comb finger electrodes.

Design and fabrication of MEMS sensors using commercial CMOS foundry services enable on-chip integration with front-end readout circuits to reduce the effects of parasitics and noises and improve the sensor performance. The integration of sensors and readout circuits on a single chip can also help reduce the overall packaging and manufacturing cost. Such integrated CMOS-MEMS sensors have been demonstrated in accelerometers (Lemkin and Boser, 1999; Jiangfeng Wu et al., 2004; Hongwei Qu et al., 2008; Tan et al., 2011; Tsai et al., 2012; Chiu et al., 2013; Chiu et al., 2014; Chiu et al., 2016; Chiu et al., 2017; Chiu et al., 2019), gyroscopes (Jiang et al., 2000; Xie and Fedder, 2001; Sun et al., 2011), pressure sensors (Jang and Yun, 2017; Liao et al., 2021), inclinometers (Chiu et al., 2015), mass sensors (Verd et al., 2008; Wei and Lu, 2012), humidity sensors (Dai, 2007; Lazarus et al., 2010) and flow sensors (Liao et al., 2013). In addition to reducing noise and amplifying signals, the on-chip circuitry can also be used to drive actuators for feedback control and self testing (Lemkin and Boser, 1999). Among these integrated sensors, accelerometers have been under intensive studies. However, tunability has not been demonstrated in integrated CMOS-MEMS accelerometers to the best of our knowledge.

This paper presents a reconfigurable/tunable z-axis accelerometer implemented with the CMOS-MEMS technology. The mechanical resonant frequency and thus the sensing bandwidth and sensitivity can be tuned by the electrostatic softening effect. The proposed sensor was designed using a commercial 0.35 μm CMOS foundry service. The tuning electrodes were implemented in the metal2 (M2) and polysilicon layers in the standard CMOS process. The sensing structure was released by post-CMOS wet metal etching. In the device design, the sub-micron tuning gap formed by removing the metal1 (M1) layer in the CMOS process allows low tuning voltage and good tunability. An on-chip chopper stabilized readout circuit was designed to convert the capacitance signal to a voltage readout and reduce the effects of parasitics and noises.

After the chip was fabricated and delivered by the foundry, post-CMOS processes were carried out to release the sensing structure. Various mechanical, electrical, and acceleration tests of the sensor are discussed in the following sections.

There are certain discrepancies between the characteristics from measurement and design. The possible causes are discussed in the following. The proposed accelerometer was designed based on the nominal process data such as the material properties and thickness of various metal and oxide layers disclosed by the CMOS manufacturing foundry. However, it is known that these BEOL thin films have residual stress that can cause the released MEMS structures to deform. The characteristics of the deformed device will then deviate from the original design. Generally speaking, devices with larger dimensions and more compliant structures, such as inertial sensors, are prone to larger deformation and related reliability problems in the presence of residual stress. On the other hand, devices with smaller dimensions or more rigid anchoring constraints, such as resonators or pressure sensors, tend to have less deformation and more reproducible characteristics. The residual stress in the deposited films in CMOS processes has been studied in the literature (Yen et al., 2011; Cheng et al., 2015; Tsai et al., 2015; Valle et al., 2017; Valle et al., 2021) but has not been well characterized and controlled in the fab. This is one of the reasons that pure monolithic CMOS MEMS devices are rare in the commercial products.

The accelerometer proposed in this study has a 315 × 315 μm² proof mass which is suspended by four springs. The tuning gap underneath the square proof mass is designed to be 0.665 μm, which is the thickness of the sacrificial M1 layer. However, as shown in Figure 9B, the gap between the release structure and the bottom tuning electrode can be as larger as 11.8 μm. Figure 14A further shows the white light interferometric measurement of the surface topology of a released device. Figure 14B is a line scan of the surface height through the center of the proof mass. It can be seen that the suspended proof mass is warped and elevated from the surface of the chip by up to about 13 μm, agreeing with the results from the FIB experiment. The raised structure and enlarged gap have several effects on the device performance:

1. The movable and fixed sensing fingers will no longer level with each other in the z axis as shown in Figure 4. Thus the sensing capacitance C ₀ can be reduced and the differential capacitance signal ±ΔC may not work as expected. This is potentially the main reason why the measured sensitivity is much smaller than the designed value.

The stress-induced deformation in CMOS MEMS devices with various BEOL metal layers and metal-oxide stacking structures have been studied in (Cheng et al., 2015; Tsai et al., 2015; Yen et al., 2011; Valle et al., 2017; Valle et al., 2021). However, it is known from the literature and our own experience that the residual stress and induced deformation is difficult to control and reproduce. The effect of stress induced deformation on the performance of capacitive CMOS-MEMS accelerometers can be alleviated by curl matching frames (Xie and Fedder, 2000), which were adopted in the proposed device. In the thermomechanical simulation in the device design, proof mass deformation similar to the measured result in Figure 14 was observed, but the deformed movable and fixed sensing fingers remained leveled to each other relatively well thanks to the matching frame. In the actual released device in Figures 9, 14, however, the height difference between the two sets of fingers are 3–5 μm, which is much higher than expected from simulation. Since the residual stress in CMOS processes is difficult to control and predict, this issue needs to be tackled from other aspects. From the structural design point of view, Tsai et al. proposed a pure-oxide design, instead of the conventional metal oxide stacking design, to reduce the thermal deformation (Tsai et al., 2015). In the current proposed accelerometer, a planar metal layer was used for the tuning electrode. Thus the pure oxide proof mass design in (Tsai et al., 2015) is not applicable. However, the metal tuning electrode in the current device can be replaced with the Ti/TiN electrode as demonstrated in (Chen et al., 2019) in the future work. The proof mass will be composed of a thick oxide layer (4–5 μm) and a much thinner Ti/TiN electrode (0.14 μm) in the bottom. Thus the oxide proof mass will remain less deformed in the presence of stress from the thin electrode layer. In addition, the thermal expansion coefficients of various materials in the BOEL layers are 0.5, 23, 8.6, 9–10 ppm/K for SiO₂, Al, Ti, and TiN, respectively. Therefore the thermal mismatch can also be reduced in the Ti/TiN electrode design compared with the current Al electrode design. Yen et al. showed thick and symmetric metal-oxide stacking realized by wet oxide etching can be used to significantly reduce the deformation in the presence of residual stress (Yen et al., 2011). Such structural design and post-CMOS processing will also be considered in future design to alleviate the stress-induced deformation. It is noted that all the suspended structures, including the proof mass, sensing fingers, and curl matching frames, need to be designed with the same layer stacking in order to compensate for any residual stress that can not be fully predicted in the design phase.

The performance of the proposed reconfigurable/tunable CMOS-MEMS capacitive accelerometer is compared with other integrated CMOS-MEMS capacitive accelerometers in the literature, as shown in Table 4. Even though the stress-induced deformation has deteriorated the performance of the proposed device, its reconfigurability/tunability is clearly demonstrated.

This paper presents a reconfigurable/tunable z-axis accelerometer based on the electrostatic spring softening effect. The proposed monolithic accelerometer and on-chip readout circuits were implemented by using a commercial 0.35 μm 2P4M CMOS foundry service. The tuning electrodes were designed in the metal and polysilicon layers in the BEOL CMOS process. The sensing structure was released by post-CMOS wet metal etching. Measurement results showed the proposed accelerometer had a sensitivity of 12 mV/g in the range of 0–5 g. The sensitivity and bandwidth tuning ranges are 50% and 18%, respectively, for an applied tuning voltage of 25 V. The measured characteristics deviated from the design values due to severe residual stress induced structural deformation.