Implement Low-Power ML at the Edge with the Avnet RASynBoard

I'm excited to introduce the Avnet RASynBoard. Avnet has partnered with Renesas, Syntiant, and Edge Impulse Studio to develop a device with software enablement targeting low-power ML at the edge.

Why is it called RASynBoard? The solution is built around a Renesas RA6M4 MCU and a Syntiant low-power always-on NDP120 Neural Decision Processor. The EVK is about the size of a Box of Rasins!

Hardware Overview

The RASynBoard is implemented with the components listed below..

• Renesas RA4M4 (Cortex-M33 @ 200MHz)
• Syntiant NDP120 Neural Decision Processor (Always-on low power neural processor)
• Renesas DA16600 Wi-Fi BlueTooth5 for wireless connectivity (Low power connectivity for BLE applications and/or cloud connectivity)
• TDK 6-Axis IMU (ICM-42671-P)
• TDK Digital Microphone (T5838)

The RASynBoard EVK consists of two boards, the RASynBoard Core board and an I/O board.

• The core board can be used without the I/O board for deploying in the field, or as an add on board for a new hardware design.
• The I/O board exposes hardware signals to common sensor interfaces such as MikroE Click Boards, and PMOD devices.

• The IMU and Digital Microphone are connected directly to the NDP120 allowing the RA6M4 and DA16600 Wi-Fi/BLE device to run in a low power state while the NDP120 process' incoming data looking for ML features
• When a feature is detected by the NDP120, an interrupt is generated to the RA6M4 to wake it up to process the event

Software andTools Enablement

The RASynBoard includes Software and Tool enablement to help you develop custom solutions quickly without having to reinvent the wheel.

Renesas Flexible Software Package (FSP)

The Renesas Flexible Software Package (FSP) is an enhanced software package designed to deliver user-friendly, scalable, high-quality software for embedded system designs using the Renesas RA family of Arm Microcontrollers.

Renesas DA16600 Software Enablement

The DA16600 module provides a convenient way to add both low power Wi-Fi and low power Bluetooth Low Energy (LE) functionality to your device. The low power Wi-Fi DA16200 system-on-chip (SoC) and the low power Bluetooth LE DA14531 SoC are integrated together on a single module. Together, they deliver long battery life and low power consumption in a convenient form factor.

• AWS, Azure, Avnet IoT Connect and generic AT CMD DA16600 images are available to jumpstart your network or cloud connected solution.

Edge Impulse

Build datasets, train models, and optimize libraries to run on any edge device, from extremely low-power MCUs to efficient Linux CPU targets and GPUs.

• Edge Impulse includes RASynBoard specific features
• Capture sensor data from the RASynBoard directly into your Edge Impulse project
• Leverage public projects to quickly import large datasets
• Design ML models targeting the RASynBoard's NDP120 neural decision processor
• Deploy ML models that run on the RASynBoard without any modifications

Avnet RASynBoard Out-of-Box GitHub Project

The RASynBoard Out-of-Box (OOB) application's goal is to provide a working example that exercises the RASynBoard hardware and gives development teams a strong starting point for their own custom designs and ML at the Edge solutions.

• Tested releases
• Configurable using a configuration file on the microSD card (no need to recompile to change application behavior)
• Supports sending inference data to Avnet IoTConnect cloud solution. (AWS and Azure support are on the roadmap)
• Drop your Edge Impulse model on the microSD card, update the configuration file and run your custom Edge Impulse ML models using the OOB application.
• In depth documentation
• Supports flashing current configuration to core board SPI Flash for core board only deployments with a button press!

What ML problems to you want to solve?

Please post comments below!

What's next?

Learn more about using the Avnet RASynBoard by checking out these other Hackster projects.