Successfully Integrating MEMS Accelerometers into High Precision Applications

Dr Panos Ioakim, Field Applications Engineer at Anglia, discusses their characteristics and gives an insight into overcoming the design challenges for the successful integrationof MEMS accelerometers into instrument grade products.

Selecting the right device
Selecting a MEMS accelerometer for an application primarily involves making decisions based on dynamic range, bandwidth of operation, current consumption, number of axes, sensitivity, and noise density. With over 50 different products currently available from Analog Devices offering a wide variety of operational characteristics and capabilities, selecting the appropriate device requires a methodical selection process.  Aiding this selection process, MEMS Inertial Sensors Selection Tables are available on the Analog Devices website, providing product segmentation by application, as is a more detailed parametric Product Selection Table, offering finer selection by attribute. Since the selection process is mostly informed by target application criteria however, arriving at an appropriate list of parts is by no means a daunting exercise. For static operations such as tilt measurement for example, the focus would typically be on devices with low bandwidth, low dynamic range and good noise and sensitivity characteristics, yielding a selection of devices such as the ADXL103, ADXL203, and the ADIS16003, to mention but a few. Worth noting is that for some applications, highly integrated solutions also exist, for example the ADIS16203 inclinometer able to provide out-of-the-box 0.025° resolution, offering a quick time to market. For high resolution specialist applications, the ADXL354 and ADXL355 offer superior performance with noise densities down to 22.5µg/ÖHz over analogue and digital outputs respectively.

For mixed dynamic and static applications such as bio-sensing wearables, prosthetics and patient monitoring where a larger dynamic range and bandwidth are required at the lowest power consumption possible, the focus would mainly be on the micro-power range of devices, such as the ADXL362 and ADXL363 which outperform in this area, offering 200Hz bandwidth and up to 8g at just 3uA operational current, whilst also incorporating 270nA motion-activated wake up and 10nA standby modes. Conversely, for applications requiring wide bandwidths and higher dynamic ranges, the ADXL100X series would be aptly suited, offering bandwidths up to 24 kHz on fast analogue outputs, whilst for lower bandwidths, the digital output ADCMXL1021-1 would suit applications to 10 kHz. For products specifically employing frequency domain analysis, more specialised modules are also available, such as the ADIS16228 which provides on-chip tri-axial FFT, storage, and a flat frequency response up to 5 kHz.

Integration into the system
Although selecting the correct device for an application based on specific performance criteria is important, integrating a chosen device successfully into a product is critical for high precision applications as it requires careful management of the many inherent characteristics of these sensors.

Fig. 1  MEMs internal inertial structure

MEMS accelerometers are predominantly based on the physical properties of the mass-spring topology, where an inertial micro-mass is suspended by a set of polysilicon springs and is free to move within limits determined by the physical geometry of the sensor, in one, two, or three dimensions (Fig. 1).  Fingers extending from the inertial mass locate between fingers fixed to the substrate in a mesh to create movable differential capacitive elements in each axis. Accelerations in the x or y directions in this example, induce a force upon the inertial mass resulting in a motion relative to the fixed substrate, thus offsetting the fingers within the mesh from their equidistant position of rest, in turn resulting in differential capacitance imbalance. This imbalance, which is proportional to the acceleration perceived by the device, is quantified via onboard circuitry to derive a proportional and stable voltage output. Due to manufacturing tolerances however, suspension spring inconsistencies result not only in a collinear to the applied acceleration vector mass displacements, but also in unwanted lateral displacements, resulting in erroneous  acceleration artefacts on the other perpendicular sensing axes. This effect is termed cross-axis sensitivity, and it is typically prominent in most sensors in the order of 1%.

Small as this effect may appear, in some applications cross-axis sensitivity can result in significant errors. For example, a sensor measuring vertical acceleration can be mathematically corrected for tilt using basic trigonometry. In reality however, the tilting of the sensor will produce asymmetrical cross-axis artefacts dependent not only on the magnitude but also the direction of tilt. In applications where displacements are derived via a dual numerical integration process, these errors can accumulate and severely distort the data. High accuracy tilt-measuring applications depending purely on trigonometric calculation of the angle and sensor sensitivity can also be affected, as can dynamic applications with sizeable cross-axial excitations.

Fig. 2  ADXL354 internal signal chain

Dealing with the output
Due to their principle of operation, the output of MEMs accelerometers is ratiometric in nature to the supply voltage of the core electronics. This ratiometricity necessitates the use of ultra low noise, highly stable power supplies for precision measurements; a requirement that can be readily met using either the AD4550 or the LTC6655 devices from Analog Devices. In sensors with internal regulators, the output is referred to the core voltage and not the supply voltage to the device, and as such, in these devices the core voltage is available on a separate pin as a reference for correct interfacing to external circuits. Further, sensor sensitivity, zero offset, and noise density are all affected not only by voltage variations but also by variations in temperature. Temperature related drift can of course be corrected by direct measurement, and many digital accelerometers offer on-chip temperature sensors for this purpose, however, the simultaneous drift of several sensor variables requires careful consideration.

fig. 3 (a) ADXL354, (b) ADXL1004 frequency response

The combined effect of internal electronic filtering and the natural resonance of the mass-spring structure results in an output frequency characteristic whose shape is determined by fabrication and is therefore specific to each product type. The contrasting frequency responses of the ADXL354 and the ADXL1004 are exemplified in Fig. 3 in which the useful frequency range can be seen to be 1 kHz and 10 kHz respectively, as indicated for the ADXL354 in Fig. 3(a) by section (i). Sections (ii) and (iii) either side of the resonant peak are generally deemed unusable, however recent research based on the ADXL325 has shown that high frequency excitation in region (iii) via the self-test pin available on most Analog Devices accelerometers, allows for signal dithering thus enabling the enhancement of low frequency signals [1].

The unique combination of microstructures and low-level signal acquisition inevitably results in intrinsic noise in MEMS accelerometers that is not only very prominent, but is by far one the most performance-limiting factors of these sensors. This inherent noise manifests itself as 1/f noise in the lower frequencies and as thermal noise at higher frequencies, consequently, for most high precision applications, the output of digital sensors necessitates further processing, whilst the output of analogue sensors requires active filtering and post-digitisation processing to obtain best possible performance. It should be noted that the internal 32kW output resistances present in the analogue sensors (Fig. 2) have very wide tolerance characteristics and they should therefore be negated by active high impedance buffering in all precision applications.

Summary
There is no question that MEMS accelerometers available to date can provide the means of delivering high precision instrument-grade products provided that in addition to correct device selection, proficient integration is also employed in order to take advantage of their notable performance characteristics, but also to simultaneously account for the effects of their inherent limitations. As boundaries continue to be challenged yielding new products annually, MEMS accelerometers are becoming more and more integrated into higher end products and will continue to do so, undoubtedly finding usage in a diversity of industries, further displacing other current technologies.

Design Support
Anglia offer comprehensive support for customer designs on the full range of Analog Devices MEMS sensors, with free evaluation kits and samples of products via the EZYsample service which is available to all registered Anglia Live account customers.

In addition, Anglia’s engineering team have extensive experience working with sensor designs and are on hand to provide valuable product demonstrations, insight, advice, and component recommendations for your application. This resource is available to support customers with all aspects of their designs, offering hands on hardware and software design support along with access to an extensive resource of technical application notes and expertise from Analog Devices.

For further information on the MEMS portfolio along with other Analog Devices products and development tools please visit www.anglia-live.com or scan the QR code.

[1]    P. Ioakim and I. F. Triantis, "On-Demand MEMS Accelerometer Dynamic Response Acquisition and Output Dithering via Self Test Pin Actuation," 2020 IEEE SENSORS, 2020, pp. 1-3