The LSM6DSOXTR is a 6-axis inertial measurement unit from STMicroelectronics that integrates a 3-axis digital accelerometer, a 3-axis digital gyroscope, and a machine learning core (MLC) into a compact 2.5 mm × 3 mm × 0.86 mm LGA-14L package. Positioned as the embedded-AI member of ST's iNEMO family, the device enables motion pattern classification directly on the sensor, relieving the main processor of the first stage of activity tracking and accelerating motion-based applications such as fitness logging, wellness monitoring, personal navigation, and fall detection. Beyond its AI capabilities, the LSM6DSOXTR provides dedicated configurable signal processing paths for optical image stabilization (OIS) and electronic image stabilization (EIS), making it equally suited to camera module applications requiring closed-loop control.

Embedded Machine Learning Core for On-Sensor Classification

The LSM6DSOXTR's machine learning core is a decision-tree-based processing engine that classifies motion data based on known patterns. The MLC works in conjunction with the integrated finite-state machine logic to handle motion pattern recognition or vibration detection. Customers creating activity-tracking products can train the core for classification using Weka, an open-source PC-based application, to generate settings and limits from sample data such as acceleration, speed, and magnetic angle that characterize the types of movements to be detected.

The MLC enables a range of classification tasks directly on the sensor, including activity recognition (stationary, walking, fast walking, jogging, biking, driving), gym activity recognition (counting bicep curls, squats, and push-ups), airplane mode detection, and stationary/motion discrimination. By executing these algorithms at the sensor level, the LSM6DSOXTR reduces the power consumption and processing burden on the host system while maintaining responsive, always-on operation.

Dedicated OIS and EIS Signal Processing Paths

The LSM6DSOXTR fully supports EIS and OIS applications through a dedicated configurable signal processing path for OIS and an auxiliary SPI interface configurable for both the gyroscope and accelerometer. The OIS functionality can be configured from the auxiliary SPI or the primary interface (SPI / I²C and MIPI I3C), providing flexibility in system architecture. This dedicated signal path enables closed-loop control for camera stabilization, allowing the sensor to provide low-latency motion data directly to camera module controllers without burdening the main application processor.

Programmable Finite State Machine

The device includes a programmable finite state machine (FSM) that supports up to 16 independent programs for custom movement recognition in low-power mode. The FSM operates on accelerometer, gyroscope, and external sensor data, enabling deterministic, low-latency event detection. Each FSM program can be configured to recognize specific gestures or motion sequences, providing a high degree of customization for application-specific motion detection requirements.

Comprehensive Digital Feature Set

The LSM6DSOXTR supports the main OS requirements, offering real, virtual, and batch sensors with 9 KB for dynamic data batching. An advanced pedometer with step detector and step counter provides accurate activity tracking. Standard interrupts include free-fall, wake-up, 6D/4D orientation, click, and double-click detection. The device also supports S4S (Sensor for Sensor) data synchronization, enabling precise time alignment between the IMU and external sensors.

Machine Learning Core Operation

The MLC operates on data from the accelerometer, gyroscope, and up to four external sensors connected through the sensor hub interface. The processing pipeline includes configurable filters, a computation block that extracts pre-defined features, and a meta-classifier that produces the final classification result based on a decision tree. The decision tree is generated offline using Weka and loaded into the sensor through the standard register interface. This workflow enables rapid development of custom classification applications without requiring deep expertise in embedded machine learning.

Sensor Hub and External Sensor Integration

The LSM6DSOXTR can acquire data from up to four external sensors through its sensor hub interface, enabling the aggregation of magnetometer, pressure, or other environmental sensor data alongside internal motion measurements. The sensor hub functionality reduces the number of separate data paths to the host processor and enables more complex sensor fusion applications. External sensor data can be processed by the MLC and FSM alongside the internal accelerometer and gyroscope data, extending the range of detectable patterns and events.

Data Synchronization and Timestamping

The device supports main processor data synchronization and S4S (Sensor for Sensor) data synchronization, ensuring precise time alignment between the IMU and external sensors or the host processor. This synchronization capability is critical in applications where multiple sensor streams must be correlated, such as camera stabilization systems where gyroscope data must be time-aligned with image frames.

The LSM6DSOXTR is distinguished from other members of the iNEMO family primarily by its embedded machine learning core and dedicated OIS/EIS signal paths. While the LSM6DSO provides similar accelerometer and gyroscope specifications without the MLC, the LSM6DSOXTR's on-sensor classification capability enables lower system power consumption and reduced host processing burden in activity tracking and motion recognition applications. The auxiliary SPI interface for OIS data output is a distinctive feature that differentiates the LSM6DSOXTR from general-purpose IMUs, making it the preferred choice for camera module stabilization applications.

Camera Module Stabilization (OIS and EIS)

The LSM6DSOXTR's dedicated OIS signal processing path and auxiliary SPI interface make it a natural fit for camera module stabilization in smartphones, action cameras, and drones. The auxiliary SPI provides a direct, low-latency data path from the gyroscope to the OIS controller, enabling fast closed-loop control of the lens or sensor position. For EIS applications, the primary interface provides motion data to the image processing pipeline for digital stabilization. The combination of OIS and EIS support in a single package simplifies camera module design and reduces component count.

Activity Tracking and Fitness Monitoring

The embedded MLC enables the LSM6DSOXTR to perform activity classification directly on the sensor, recognizing stationary, walking, fast walking, jogging, biking, and driving states without host intervention. This capability is particularly valuable in fitness trackers, smartwatches, and health monitoring devices where battery life is critical and continuous host processing is impractical. The hardware pedometer provides accurate step counting, while the FSM can recognize specific gym activities such as bicep curls, squats, and push-ups.

Wearable Devices and Hearables

Wearables and hearables demand always-on motion sensing with minimal power consumption and a small footprint. The LSM6DSOXTR's 0.55 mA combo high-performance current and 3 µA power-down current enable extended battery life, while the 2.5 mm × 3.0 mm × 0.86 mm package fits easily into space-constrained designs. The embedded MLC and FSM handle motion recognition tasks at the sensor level, allowing the host processor to remain in a low-power state during periods of low user activity.

IoT Connected Devices and Personal Navigation

The LSM6DSOXTR supports IoT connected devices and personal navigation applications through its sensor hub functionality, low power consumption, and embedded processing capabilities. The device can aggregate data from external magnetometers and pressure sensors to provide heading and altitude information for pedestrian navigation. The MLC can detect context changes such as airplane mode activation or stationary/motion transitions, enabling context-aware power management and user interface adaptation.

Vibration Monitoring and Industrial Sensing

The machine learning core's ability to classify vibration patterns enables the LSM6DSOXTR to be used for vibration monitoring in industrial equipment and consumer devices. The FSM and MLC work together to recognize vibration signatures associated with specific operating conditions, enabling on-sensor anomaly detection without continuous data streaming to a central processor.

STMicroelectronics provides a comprehensive ecosystem of development tools for the LSM6DSOXTR. The STEVAL-MKI109V3 professional MEMS tools board serves as the motherboard for adapter boards, enabling connection to a PC for configuration and data logging. The STEVAL-MKI197V1 is the dedicated adapter board for the LSM6DSOX, providing the complete pinout for evaluation in the user's application. The X-NUCLEO-IKS01A2 expansion board for STM32 Nucleo provides a platform for rapid prototyping of motion sensing applications, with support for the LSM6DSOX through the standard STM32 development environment.

For machine learning development, ST provides the Unico-GUI graphical interface and AlgoBuilder application for the graphical design of algorithms and decision trees. The FP-AI-CTXAWARE1 function pack demonstrates human activity recognition using the LSM6DSOX machine learning core and neural networks running on an STM32 microcontroller, providing a reference implementation for context-aware applications. Arduino library support and platform-independent drivers further lower the barrier to entry for developers working with the LSM6DSOXTR.

The LSM6DSOXTR represents a significant step forward in motion sensing, transforming the inertial measurement unit from a raw data provider into an intelligent classification engine. Its embedded machine learning core, programmable finite state machines, and dedicated OIS/EIS signal paths address the power, latency, and processing constraints of battery-powered devices and camera modules. By enabling motion pattern recognition and activity classification directly on the sensor, the LSM6DSOXTR reduces host processor burden, extends battery life, and accelerates the development of context-aware applications. For engineers designing wearables, camera systems, IoT devices, and fitness equipment, the LSM6DSOXTR offers a well-supported, production-ready solution that brings edge intelligence to motion sensing.