Algorithms and software frameworks drive new innovations in MEMS sensors for tiny wearable devices vying for less power usage and more AI features

By Majeed Ahmad, contributing writer

Wearable devices
are a hotbed of sensors as well as a design conundrum due to size, power, and
integration challenges. At the same time, however, major advancements in sensor
and microelectromechanical (MEMS) technologies are reinvigorating the analog
side with more accurate measurements and the digital front with more robust
sensor fusion.

For a start,
sensor miniaturization is a key driver for space-constrained wearable designs.
For example, motion sensors are now available in tiny packages of 2 × 2 mm for
fitness trackers, smartwatches, and hearables. And these compact sensors come
integrated with plug-and-play functions like a step counter, so they can be
easily incorporated even into clothes, jewelry, and watches.

Take the example of the ICM-20648, a six-axis motion tracking chip from TDK’s InvenSense operation, which comes integrated with a motion-sensing processor and software algorithm. That allows the wearable sensor IC to offer features such as a built-in activity classifier, calorie counter, and bring-to-see gesture tuned for wrist-worn wearables.

Fig. 1: Wearable devices like smartwatches will add more sensors
to boost functionality and utility. (Image: TDK)

also apparent is that wearables are commonly associated with activity-tracking
devices such as fitness bands and sports watches. But many of these devices
are so much more than simple monitoring gadgets. So there
is a need for new types of sensors that can bring new functionalities to
wearable devices. Take environment sensors, for instance, that can employ
mapping applications to allow wearable devices to check the air quality and,
thus, identify most polluted areas in a town.

Beyond fitness trackers
The demand for basic
wristbands is waning,
and at the same time, the market for wearable electronics in health care is
quickly expanding, especially for personal health-monitoring devices such as smart
patches and digital blood pressure monitors.

These wearable
devices can monitor heart
rate variability, oxygen levels, cardiac health, blood pressure, hemoglobin,
glucose, and body temperature. Here, more sophisticated algorithms combined
with sleek new sensors ensure that these health-care wearables are far more accurate
for many medical use cases.

These types of medical wearables are not only
a growing market; their focus on higher accuracy and greater performance also makes
them less prone to price pressures that designers commonly face in consumer
wearables. However, these medical designs face the challenge of long
qualification periods. Here, validated and certified sensor modules come into

Take the case of
a reference design
for wearable
devices providing 24/7 cuffless blood pressure measurement
by the Premstaetten,
Austria-based sensor supplier ams. It’s built around the AS7024 sensor chip and
software that conducts blood pressure measurement, heart rate measurement
(HRM), heart rate variability (HRV) monitoring, and electrocardiograms (ECG).

The AS7024 sensor chip includes three
LEDs, photodiodes, an optical front end and sequencer for HRM, and an analog
front end for ECG, a standard method for measuring the electrical pulses
generated by the sinoatrial node. The sensor chip’s HRM operation is based on
photoplethysmography (PPG), a technique that uses sampling light to measure the
pulse rate in blood vessels, which expand and contract as blood pulses through

The accompanying
software in this reference design analyzes the synchronized HRM and ECG
measurements to calculate blood pressure. According to ams, the AS7024’s blood
pressure measurements have been validated in a clinical trial at the Medical
University of Graz in Austria as per IEEE standard for cuffless wearable

Wearables’ power crunch
Power consumption,
a crucial issue in portable electronics, becomes even more critical in wearable
devices that mostly offer always-on detection and measurement functions. Moreover,
wearable devices usually feature smaller-capacity batteries due to size
constraints. Consequently, for sensing designs in wearable devices, engineers must
choose between low power and high performance.

Therefore, sensor
makers are now employing innovating new techniques to further lower the power
draw from always-on sensors without performance and accuracy trade-offs. A new
crop of solutions combines MEMS sensors with algorithms and firmware that
intelligently process, synthesize, and calibrate the output of sensors.

A new position
tracking sensor from Bosch Sensortec is a case in point. The BHI160BP sensor employs
an algorithm for
pedestrian dead reckoning (PDR) to calculate the user’s relative location based
on data collected from the inertial sensor. Then it re-calibrates itself every few minutes
to obtain the absolute position provided by the GNSS/GPS module.

In other words, the
GNSS/GPS module,
which can rapidly drain a device’s battery capacity, is kept in sleep mode for
most of the time. That allows users to navigate reliably and extends GPS
tracking in wearable devices from several hours up to several days.


Fig. 2: The always-on position tracking sensor is optimized for
use with GPS/GNSS modules. (Image: Bosch Sensortec)

Another MEMS
sensor from Bosch, the BMA400, claims to draw 10
times less current than existing accelerometers due to intelligent features like built-in
activity recognition. It wakes up automatically only when it
detects motion and goes back to sleep mode when the motion stops. The BMA400 accelerometer consumes 14 µA at the highest performance and falls to
1 µA and below in the ultra-low-power self-wake-up mode.

This acceleration
sensor for wearable devices handles continuous measurement by using precisely
defined cut-off frequencies. That, in turn, makes motion sensors resistant to
vibrations, so wearable devices can distinguish between real alarm situations such
as broken glass and false signals coming from random noise and vibrations.

Wearable sensors meet AI
What really makes
wearables smart is artificial intelligence (AI), and that’s partly enabled by MEMS
sensors featuring built-in local intelligence, which comes in the form of algorithms
and software frameworks. These software solutions combined with hardware
accelerators incorporate deep learning and other AI features to contextualize
individual behaviors and surroundings. They independently process sensor data, and
that also reduces power consumption, lowers cost, and boosts overall efficiency.

For example, the BMA400
accelerometer from Bosch
mentioned in the above section integrates the motion classification
functions. Bosch also offers smart sensor hubs, such as the BHI260 and BHA260 devices,
which integrate MEMS sensors with a microcontroller for low-power sensor data
processing and data buffering alongside a software framework that includes sensor fusion.

These low-power solutions incorporate data
from multiple sensors and enable pre-processing for always-on execution. They facilitate
features such as 3D orientation, step
counting, position tracking, activity recognition, pose-and-head tracking, and
context awareness in wrist-mounted products, hearables, eyewear, and other
wearable devices.

surprisingly, therefore, software assets are becoming vital in the
arsenal of MEMS sensor suppliers. It’s the
algorithms and software frameworks that allow MEMS sensors to offer an accurate live classification of AI
data and enable features like activity tracking without utilizing precious
processor resources. It’s also a practical manifestation of the edge computing
that has mostly been discussed in the context of IoT

designers are building a whole new industry one gadget at a time, and MEMS
sensors are an intrinsic part of this design movement. Wearable designs have
come a long way from counting steps in fitness trackers, and they are already
applying machine-learning algorithms to classify and analyze data.

MEMS sensors are enabling this without requiring an external microcontroller,
which cuts cost, lowers power usage, and reduces design complexity. Next, the
symbiotic relationship between MEMS sensors and AI is expected to take wearable
devices into a whole new space.