The increasing complexity of low-end devices is giving rise to a new form of data processing. Instead of a main controller performing all the heavy lifting, attached peripherals can have integrated controllers and DPSs which can pre-process their own data before passing along to the main controller which can offload work from the main controller, potentially saving power. An example of this is the LSM6DSOX, ST's latest inertial module that has machine learning capabilities.
The LSM6DSOX is an always-on 3D accelerometer and 3D gyroscope. The sensor can be accessed via SPI and I2C, as well as the industry's latest standard, I3C. The sensor is compliant with Android for integration with Android systems (such as tablets and phones).
It also incorporates significant motion detection and tilt detection. The tilt motion built into the sensor is capable of triggering events during tilt changes. For example, an event can be triggered when a user has their phone in their pocket and stand up after having been sat down. This tilt detection is done all in hardware without the need for a controller, which can greatly reduce power and offload work from a processor.
Machine learning core in the LSM6DSOX
However, the real magic behind the LSM6DSOX is the internal machine learning system. The machine learning core, realized by a series of “if-then-else” conditions, allows the identification of specific tasks such as walking, running, and driving which are user programmable.
With up to 256 nodes available, the machine learning core can execute up to eight flows simultaneously and each flow can generate up to 16 results. The sensor can also take external sensors into account with the use of the Sense Hub whereby other sensors can send their data to the LSM6DSOX.
External sensor connection to the LSM6DSOX
"Machine learning is already used for fast and efficient pattern recognition in social media, financial modeling, or autonomous driving," said Andrea Onetti, STMicro's Group Vice President of Analog, MEMS, and Sensors. "The LSM6DSOX motion sensor integrates machine-learning capabilities to enhance activity tracking in smartphones and wearables."
- 2.5mm x 3mm x 0.83mm package
- Power consumption of 0.55mA
- FIFO of up to 9KB
- ±2/±4/±8/±16 g full scale
- ±125/±250/±500/±1000/±2000 dps full scale
Integrating small, low-powered controllers into peripherals usher in a shift in the way that hardware is designed. By offloading as much work as possible from a main processor (which may not be the most power conserving hardware), devices could see higher through-put with pre-processed data as well as longer battery life.
- Sensor Fusion for Autonomous Vehicles: Interview with Tactile Mobility CEO Amit Nisenbaum
- Bosch Introduces Position Tracking Smart Sensor as Part of Third Wave of Sensor Technologies
- ams Introduces Image Sensor for High-Throughput Manufacturing and Optical Sensing Applications
- Achieving Angle of Light Detection: Silicon Nanowires Emulate a Gecko’s Ears