SmartEdgeML™

Redefine your Edge ML experience

SmartEdgeML: Run Machine Learning on 6-axis motion sensors

TDK’s SmartEdgeML is an advanced solution that leverages the combined power of TDK’s SmartMotion sensors and machine learning to create a new paradigm of Edge ML, which enables next-generation applications for smartwatches, hearables, health monitors and more.

TDK’s SmartEdgeML solution allows developers to implement ML algorithms on a small IMU sensor chip, offloading tasks from the Edge processors, thereby reducing the amount of data going to the Edge processor. This significantly improves battery life, data privacy, and Edge processor computation resources.

1. Get the SmartBug 2.0 multi-sensor wireless hardware. It contains the ICM-45686-S IMU
2. Collect data from SmartBug 2.0 wirelessly and input it into the SIF tool.
3. Train Phase: SIF will build an optimized ML model (algorithm) based on the input training data, features and parameters.
4. Run/Test Phase: Run the model on the ICM-45686-S IMU. Test the algorithm with SmartBug 2.0.
5. If the evaluation is satisfactory, use the SIF Integration guide to run the ML-based algorithm on the chip on your system.

• Auto Mode: Allows a beginner to build a ML model in 2 minutes. All you have to do is collect data and the SIF tool will take care of the rest.
• Ultra-low power: Build ML models at < 30 µA.
• Optimized: Run ML models on an IMU chip as small as 2.5 x 3 mm.
• Easy Visualization: View performance reports, chip utilization metrics, 3D feature data/classes, filter configurations etc.

Sensor Users: Traditionally sensor on-chip SW was fixed and offered less flexibility to implement new features on sensors. SmartEdgeML enables the following:

• Customization: User can define their own use-case and build an algorithm with SmartEdgeML in less than 2 minutes (with AUTO mode). User can also define their custom sensor settings, filters, features as inputs to the ML model.
• Ownership: User can own the data and test the ML model with their own dataset rather than depending on the sensor vendor.
• Time to market: SmartEdgeML provides both the hardware and the framework with an evaluation system such as SmartBug 2.0, to build, test, tune, and evaluate ML models quickly. This removes dependency on the support needed from the sensor vendor to build and tune sensor algorithms. In addition, there are ready examples in SIF that allow a user to quickly understand and evaluate a ML solution.
• Better battery life: With SmartEdgeML, a user can now find out directly from an IMU sensor whether he is sitting, standing, walking, or running. He can further find out if he was doing a bicep curl or a deadlift. This allows the Edge devices to sleep longer and process only smart data coming from the sensors, reducing battery drainage and MIPS cycles.

Wearables

Hearables

Laptops/Phones

Appliances

• Run vs walk vs Sit vs Stand vs in-car
• Shake Wrist vs Flick vs Slice vs Turn
• Jumping Jacks vs Bicep Curl vs Knees

• Head movement – Nodding Yes or No or Nothing

• In pocket vs out of pocket

• Coffee maker – Coffee tank empty or not