AI/ML Solutions for Edge-Enabled Product Development

[Aldrina_B]

Hi Team,

I am currently exploring AI/ML solutions for our edge-enabled product development projects, particularly focusing on wearable devices with small footprint SoCs (flash ~1 MB, RAM: 48kb - 256 kb). Our use cases involve collecting usage information, developing AI/ML models, converting these models to .h/.c format for firmware integration, and deploying them to MCUs via OTA updates. The MCUs will then collect sensor/parameter data and send it back to our AI/ML engine for continuous analysis and accuracy improvement.

Given our device footprint requirements (small feature size and cost efficiency), we are seeking information on the following:

1. Chipsets Supporting Edge AI Capability:

   • Which chipsets do you offer that support edge AI capabilities within the specified memory constraints?

2. Toolchain Support for AI/ML Model Generation and Conversion:

   • What toolchains do you provide for generating AI/ML models and converting them to formats suitable for firmware integration?

3. Toolchain Support for Firmware Development:

   • What toolchains are available for firmware development to enable AI/ML and edge AI capabilities on your chipsets?

4. Application Support:

   • Do you offer active and prompt application support during the project to help implement solutions and address knowledge gaps?

5. Alternative Solutions:

   • Are there any alternative solutions or approaches you recommend to overcome potential challenges or blocks in implementing AI/ML on small footprint devices?

Thanks in advance

[Øyvind]

Hello, 

Yes, Nordic Semiconductor's support is centered around the Edge Impulse wrapper. The usage of the wrapper is demonstrated by the Edge Impulse: Wrapper sample, while the nRF Machine Learning application demonstrates the complete out of the box reference design. There is also our own Edge Impulse mobile app developed in collaboration with Edge Impulse to train and deploy embedded machine learning models on the Nordic Thingy:53 using Edge Impulse Studio.

In the Edge Impulse: Wrapper sample you can see what development kit or devices that are supported. 

Kind regards,
Øyvind