Abstract
This paper presents a novel hybrid CMOS/MEMS tilt sensor with a resolution over a range. The device uses a MEMS-based semicircular mass suspended from a rigid body, projecting a shadow onto the CMOS-based optical sensor surface. A one-dimensional photodiode array arranged as a uniformly segmented ring is then used to determine the tilt angle by detecting the position of the semicircular mass. The complete sensor occupies an area of under 2.5 mm 2.5 mm.
1. Introduction
An increasing number of medical devices including implantable prosthetics and body worn instrumentation are incorporating sense systems within and around the body. Physical constraints demand such devices to be compact and lightweight, and the need for autonomy imposes stringent power budgets on such systems. One such sensor is the inclinometer (or tilt sensor), which senses its orientation with respect to gravity.
Sensing tilt, in personalised medicine is becoming increasingly important, especially when combined with other sensor modalities. For example, a tilt sensor incorporated within a heart monitor can provide a metric on the patients’ activity, providing additional insights for interpreting the electrocardiogram data. In other biomedical applications, real-time, accurate tilt data can be used in therapy, for example, in correcting vestibular balance disorders.
The vast majority of tilt sensors developed to date are capacitive (or conductive) in nature, often with electrolytic fluid-filled cavities. Such systems have been developed by Bretterklieber et al. [1] and more recently using MEMS technology [2, 3]. Zhao and Yeatman [4] have developed a digital MEMS-based tilt sensor offering the possibility of simplifying the interface electronics. It has also been reported possible to implement inclinometers using modified accelerometers [5, 6] but such systems are typically limited in range and/or sensitivity. Other approaches include a MEMS convection-based [7] and optical microfilter-based [8] tilt sensors.
In this paper, we present a novel ultra-low-power tilt sensor generating a digital output. This has been implemented as part of a hybrid two-chip solution (MEMS microstructure/CMOS vision chip) and has been fabricated.
2. System Architecture and Circuit Implementation
This tilt sensor uses a micromechanical semicircular mass to express the gravitational field optically and then custom CMOS vision chip to detect and resolve the vector. This concept is illustrated in Figure 1.
The design splits the 330° field of view into two 165° half-planes, which we have chosen to refer to as left- and right-hand-side (LHS, RHS) horizons. This is illustrated in the top-level system architecture, shown in Figure 2. As the micromechanical semicircular mass moves, it forms a shadow that is projected onto the surface of the CMOS chip, the edge of which crosses both the LHS and RHS horizons. Therefore, each horizon detector is required to sense the position (or tilt angle) of these edges. By taking the difference, the output can encode the tilt angle using a signed, digital representation.
3. Horizon Detector Circuit
The circuit implementation of the half-plane horizon detectors is shown in Figure 3. This consists of 33 photodiode elements positioned equidistantly around an arc to detect the position of the semicircular mass. Between each two adjacent photodiodes, we use a discrete edge detector element (that is in total, 32 edge detectors per horizon) to compare photocurrent magnitudes and if sufficiently different flag an edge-detect signal. The output is captured using a sequence of operations. Firstly, the address bus is initialised using pull-up devices to precharge the parasitic capacitance. Next, a DRIVE signal is propagated through the one-dimensional array, that is, elements, until an edge signal is encountered. On detecting an edge condition, the current position is asserted onto the address bus via the address encoder, constituting (for each element) the relevant pull-down switches. Subsequently, the DRIVE signal is terminated and the address bus is sampled onto a 5-bit register.
4. Edge Detector Circuit
Figure 4 shows the circuit schematic of the edge detector element. This has been adapted from Constandinou et al. [9, 10] to the expected contrast ratio and photocurrent levels. A global tuning mechanism is maintained by driving the and references off-chip and internally duplicating and distributing to the individual edge detector elements. The DRIVEIN and DRIVEOUT are used to test an individual edge detectors state before propagating to test the next element. Additionally, an artificial propagation delay is implemented within the output buffer to guarantee robustness.
5. Photodiode Design
The photodiodes are implemented using an n-well/p-substrate junction as illustrated in Figure 5. The different photodiode elements (66 devices) have been individually custom designed (to comply with design rule geometry) to fit within a 550 μm to 600 μm radius arc in 5° segments. The active sensor surface for each photodiode element comes to: 4245 μm2. The photodiode is reverse-biased by stacking two diode-connected PMOS devices to Vdd. The photocurrent range for the given device under the expected light levels is from 1 pA (dark current) to 100 nA. The photodiode achieves a spectral responsivity of 0.15 A/W, 0.25 A/W, and 0.3 A/W for wavengths nm (blue), nm (green), and nm (red), respectively, when being uniformly illuminated at 5 μW/mm2.
(a)
(b)
6. 4-Phase Clock Generation Circuit
The operation of the tilt sensor is centrally regulated via a 4-phase clock sequence, generated using the circuit shown in Figure 6. This implements a looping 4-element state machine to generate the required control signals. Phase actions are: to precharge the bus, to interrogate the LHS and RHS horizons and drive the bus, to sample the bus result on horizon register, and to sample the difference (i.e., subtraction) on the output register.
7. CMOS/MEMS Integrated Circuits
Figures 7 and 8 show the CMOS and MEMS microphotographs, respectively.
The MEMS component has been designed such that the pivot/axel used to suspend the mass is smooth and has little friction. In addition, we patterned the release holes in the semicircular mass with a specific pattern recommended by the Tronics foundry, which forms bumps which ensure that there is enough space between the main surface area and the sidewall to avoid stiction [11]. Furthermore, the mass includes antisticking holes to ensure release, based on technical data provided from Tronics foundry. During fabrication, the design incorporates a number of struts to secure the mass during processing and release. After manufacture, these struts have been intentionally designed to be breakable, thus releasing the semicircular mass into operation.
8. Simulation Results
The circuit was simulated using the Cadence Spectre (5.1.41isr1) simulator with foundry supplied BSIM3v3 models. Parametric operating point simulations have been performed to determine the onset of edge detection for various bias conditions. These results are illustrated in Figure 9. The family of curves (for different values of ) illustrates the minimum contrast ratio () required to flag an edge detection condition. The tunability (for different values of ) is represented in the -axis variation. These results show main trends of operation: (i) for the tunable range is not significantly effected by bias current level, and (ii) for , increasing the bias current level significantly increases the tunable range (in particular extending the contrast ratio). It therefore envisaged that if the nonuniformity of incident light in addition to any component mismatch due to technology variation is maintained to be within the tunable limits, then relatively low bias current levels can be used (i.e., to 10 nA).
A top-level transient simulation illustrates the system-level operation of the tilt sensor, presented in Figure 10. This simulation assumes an average photocurrent of 25 nA, an ambient off-current of 1 nA, and uses a system clock of 500 kHz. The total power consumption is 33 μW.
The functioning of the system can easily be traced by following the transient simulation results shown in Figure 10 as follows: (i) on the rising edge of , the two buses are precharged (). (ii) On the rising edge of , the horizon detector results are asserted onto the buses (results of 9 and 14 for LHS and RHS, resp.). (iii) On the rising edge of , the values on the buses are sampled to the horizon output registers and finally, (iv) on the rising edge of , the subtracted result (from asynchronous adder) is sampled on the output register.
9. Discussion
One aspect of the sensor that has been omitted from the scope of this p is the light source. To ensure robustness in dark and fluctuating light conditions, the sensor is envisaged to incorporate a dedicated light source. This can be a small area, focused (i.e., narrow viewing area) light-emitting diode (LED) to stimulate the photodetectors within the vision chip. The composite emitter/sensor can then be packaged (e.g., as in optoisolators) to seal the components from environmental exposure.
Assuming a typically red (GaAsP) LED has a forward bias of 2 V and the current is limited to 1 mA, the emitter power consumption is 2 mW. For a 2 mm2 area emitter with 1 mm emitter/sensor separation, a 15° beam divergence and 5% luminous efficiency, this equates to 28 μW/mm2 incident light power on the CMOS surface. Therefore, for a photodiode responsivity of approximately 0.3 A/W (at nm), and photodiode area of 4245 μm2, the generated photocurrent in each illuminated segment will be 36 nA, well within the operating range of the edge detector [9].
To minimise power consumption, the light source can be biased at reduced intensity as the photodetectors cover a large sense area and is therefore extremely efficient. Alternatively the sensor can be chosen to operate to a duty cycle, that is, pulsing the emitter in synchronisation with the clock, and choosing to power the emitter only at the required refresh rate. For example, if only a 10 Hz tilt response is required, the sensor can be configured to operate for a single acquisition, that is, for four clock cycles (400 microseconds using a 10 kHz clock), every 100 milliseconds, thus reducing power consumption by 99.6%. Fundamentally, the minimum power consumption is only limited by: (i) the dark current of the photodiodes, (ii) the minimum detectable contrast ratio (for reliable edge detection), and (iii) the required sensory response time.
10. Conclusion
In this paper, we have presented a hybrid CMOS/MEMS inclinometer that uses a photodetector array to detect the position of a suspended microsystem. With a uniform array of radially positioned photodiodes, the tilt sensor achieves a 5° resolution over a 330° range. The system power consumption depends highly on the incident light intensity but expected to be within the range 10–35 μW. The target system specifications are given in Table 1.
Acknowledgment
This work was supported by the Cyprus Research Promotion Foundation (RPF) Grant no. E-0505/07.