TDK launches SmartEdgeMLTM to enable ultra-low power machine learning models on 6-axis IMU

MEMS Sensors

TDK launches SmartEdgeMLTM to enable ultra-low power machine learning models on 6-axis IMU

• The SmartEdgeML solution allows users to build, test, debug and deploy machine learning (ML) models on sensor chips

• SmartEdgeML includes: SmartMotion™ ICM-45686-S 6-axis motion sensor, SmartBug 2.0 evaluation kit, and Sensor Inference Framework (SIF) software (downloadable)

• SmartBug 2.0 (MD-45688-ML) released in February 2024

TDK Corporation has launched the InvenSense SmartEdgeMLTM solution, an advanced edge machine learning solution that provides users with new opportunities to run machine learning (ML) models on sensor chips in wearables, hearables, augmented reality glasses, Internet of Things (IoT), etc. SmartEdgeML is the first solution to successfully generate and run machine learning models on a 6-axis motion sensor IMU with a size of 2.5 x 3 mm (power consumption <30 µA).

“TDK’s SmartEdgeML will bring a paradigm shift in the edge machine learning market as it will allow developers, original design manufacturers (ODMs) and original equipment manufacturers (OEMs) to implement fully optimized motion sensing algorithms through machine learning on just a single IMU sensor chip. This will significantly reduce the amount of raw data transmitted to the edge processor, thereby extending device battery life, protecting data privacy and reducing system latency,” said Sahil Choudhary, Director of the Motion Sensor and Software Business Unit at InvenSense, a TDK Corporation company.

TDK also announced the availability of the InvenSense SmartBug 2.0 (MD-45686-ML), a multi-sensor wireless module consisting of the InvenSense ICM-45686-S IMU, which serves as an ideal evaluation system for users to get started with the InvenSense SIF machine learning software and the ICM-45686-S IMU.

SmartEdgeML consists of the following three parts:

●SIF (Sensor Inference Framework) software: SIF is the software component of SmartEdgeML, a complete machine learning framework provided by TDK, providing users with a one-stop solution to collect IMU sensor data, select custom functions, build machine learning models, test machine learning models, and deploy and run these models on the ICM-45686-S IMU through SmartBug 2.0. Tested examples include various algorithms such as motion classification (squats, jumping jacks, lateral raises, or push-ups) and wrist gesture classification (fighting, turning, shaking, or stillness).

●ICM-45686-S IMU: This is the core sensor chip of SmartEdgeML. SmartMotion ICM-45686-S is a 2.5 x 3 mm TDK BalancedGyro™ series IMU that allows machine learning decision tree models to run inside the sensor with minimal power consumption (<30µA). This new IMU has excellent temperature stability and vibration suppression capabilities, making it ideal for applications that require high-performance and ultra-low-power machine learning algorithms, such as augmented reality and virtual reality glasses, optical image stabilization applications, drones, TWS headphones, and robots.

●SmartBug 2.0 (machine learning version): MD-45686-ML This is the core hardware platform of SmartEdgeML. It is equipped with ICM-45686-S 6-axis motion sensor and is compatible with SIF. The small size and BLE + USB interface of SmartBug 2.0 allow users to quickly get started with SIF and easily use all functions from data collection to building ML models and then deploying to ICM-45686-S IMU. It can be called a "magic weapon" for getting started with SmartEdgeML.

In less than five minutes, you can learn how to create a machine learning algorithm on a motion sensor that measures 2.5*3 mm.

the term

• Machine Learning at the Edge: Machine Learning Running on Sensor Chips

• IMU: Inertial Measurement Unit

• IoT: Internet of Things

• SIF: Sensor Inference Framework

• TWS: True Wireless Stereo

Main Applications

• Hearable devices

• Wearable devices

• Augmented reality glasses

• Internet of Things (IoT)

Key Features and Benefits

• Function Customization: Users can define and customize their preferred application scenarios and build motion sensor algorithms using SIF in less than five minutes (using AUTO mode). Users can also configure custom sensor settings, filters, and functions based on their sensor algorithm requirements.

• Data privacy: Users retain their data and use these datasets to test ML models, without having to rely on sensor vendors to collect data.

• Rapid prototyping: SIF AUTO mode allows beginners to learn to create ML models in five minutes. After the user collects data, SIF automatically completes the subsequent steps. Once the model/algorithm is ready and meets the performance standards, TDK will provide an integration guide to successfully run the final algorithm on the ICM-45686-S IMU sensor in the user's system. This end-to-end ML solution based on IMU sensors saves months of algorithm work.

• 超低功耗：SmartEdgeML解决方案的功耗可低至<30 µA。低功耗能延长边缘处理器设备的休眠时间，而且只需处理自传感器的智能数据，显著延长了电池的续航时间和MIPS周期。

Key data