Fitness Coach on a Chip

This tiny chip is truly an all-in-one package, combining an IMU, AI processor, and several built-in machine learning algorithms.

Nick Bild
1 year agoMachine Learning & AI
Smart sensor system with built-in machine learning algorithms (📷: Bosch Sensortec)

For many applications, especially in wearable devices, the closer that AI processing can take place with respect to the sensor that is feeding it input data, the better. Traditional architectures require that sensor data be transferred to either a separate, on-device processing unit or remote, cloud computing resources before the machine learning algorithm can run and provide an inference. Naturally, this movement of data results in delays being added to the AI processing pipeline, from factors like transferring data in and out of system memory to network latency. These extra processing steps, and additional computational resources, also require more energy to do their work — and that is bad news unless you happen to enjoy frequently recharging your wearable devices.

For these reasons, advances in processor technologies, and their miniaturization, are increasingly being leveraged to build smart sensors that can run machine learning algorithms immediately on the data that they collect, as they collect it. This saves the time and energy that would normally be wasted transferring sensor measurements to external processing units. An announcement by Bosch Sensortec at the recent Consumer Electronics Show is the latest step forward in the AI-on-sensor market, and looks very promising from a technical perspective, but the really exciting part of their announcement is the built-in, ready-to-use software algorithms.

Specifically, they announced the BHI360 intelligent inertial measurement unit (IMU), and the closely related BHI380. The tiny 3 mm x 2.5 mm package houses a super-sensitive accelerometer and gyroscope. This sensing capability is paired with an integrated, low-power custom 32-bit processor that can run machine learning algorithms in real-time as data is collected. And as previously mentioned, a variety of useful algorithms come preloaded — the BHI360 has onboard software for simple gesture recognition and personalized sound experiences. The BHI380 chip also provides algorithms that are suitable for fitness tracking, pedestrian navigation, human-machine interaction, and more.

The new chips are the smallest programmable dual-core IMU sensors currently in the market. And with a current consumption of less than 600 μA for many use cases, they look like a great fit for long lasting wearable device designs. When the included intelligent algorithms match one’s use case, these smart sensors will help device construction move quickly, but when that is not the case, there are multiple SPI, I2C, and GPIO interfaces for integrating with external resources.

Bosch Sensortec used their new technology to demonstrate how fitness training and tracking, and personalization of workouts, can be achieved using the on-board algorithms. Using the additional features of the BHI380, they also showed how a swim tracker that measures swimming style and helps athletes to improve their technique can be created. Showing how the chips can easily integrate with existing devices, they also explained that the BHI380 can run a pedestrian dead reckoning algorithm that will keep a person on the right track even when their GPS signal drops out for a few minutes.

More information about the new smart sensor system is available on the Bosch Sensortec website.

Nick Bild
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles